Liquid-cooled component mounting plate

By using a die-based liquid-cooled heat transfer plate and a highly conductive diffuser in the data center, the heat dissipation problem of high-density computing components was solved, achieving efficient and reliable thermal management and cooling effects.

CN123075853APending Publication Date: 2026-11-10NVIDIA CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610626769.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2025-11-05
Filing Date
2026-05-08
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Thermal management challenges for high-density computing components in data centers, especially the heat dissipation of high-power chips, are caused by insufficient thermal conductivity of existing spacer pads, leading to heat accumulation and affecting component reliability and processing power.

Method used

It employs a core-based liquid-cooled heat transfer plate (TTP), combined with a highly conductive diffuser and copper tube loop, and achieves efficient heat transfer and uniform cooling through gap pads and coolant flow rate control.

Benefits of technology

It improves the heat dissipation efficiency of high-density computing components, reduces voltage drop risk, reduces the number of mounting holes, enhances reliability and design robustness, optimizes coolant flow rate, and ensures uniform cooling of each chip.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123075853A_ABST
    Figure CN123075853A_ABST
Patent Text Reader

Abstract

This disclosure relates to a liquid-cooled component mounting plate. A system for cooling a computing component includes a substrate supporting one or more processors and a heat transfer plate. The heat transfer plate includes one or more cutouts for accommodating and receiving one or more processors. The heat transfer plate is configured to remove heat from one or more processors.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Cross-reference to related applications This application claims the benefit and priority of U.S. Provisional Application Serial No. 63 / 802,043, filed May 8, 2025, entitled “THERMAL TRANSFER PLATE (TTP) FOR DIE REFERENCED COOLING”, filed pursuant to 35 USC § 119, the entire disclosure of which, for all purposes and in all its teachings, is incorporated herein by reference in its entirety. Technical Field

[0002] The embodiments of this disclosure generally relate to thermal solutions for computing hardware, and more specifically to cooling components for data centers. Background Technology

[0003] As data centers continue to operate at higher speeds (e.g., 100Gb and above) and semiconductor dies become larger in response to increasing processing demands, the thermal requirements for data center components also increase. Heat generation in data centers is primarily caused by the electricity consumed by servers, storage devices, and other hardware components. Almost all of the energy consumed is converted into heat as a byproduct of data processing, essentially meaning that for every watt of electricity used, one watt of heat is generated. Heat management is crucial; otherwise, data center components can fail under extreme thermal loads. For example, processing units may generate heat during operation, which can lead to decreased processing capacity and, in some cases, failure of these processing units and connected system components. Summary of the Invention

[0004] Boards supporting designs with one or more processors, modules, and PCIe cards are becoming increasingly dense (e.g., more components are mounted on the board, reducing available space). Increased density correspondingly increases the overall power generated on the board, modules, and / or cards. Dense backplanes and modules do not have enough space to set up separate mounting holes for using individual cold plates, die-based, corresponding to each high-power chip.

[0005] One solution to the problems mentioned above is to use a common board-based cold plate or wet heat transfer plate (TTP) for multiple chips using spacer pads (e.g., thermal gap pads that correspond to a flexible, conformable material used to fill the air gaps between heat-generating electronic components and one or more heat sinks). This presents significant thermal challenges for high-power chips due to the relatively low thermal conductivity of the spacer pads and the relatively high power density of the chips.

[0006] The embodiments of this disclosure present another improvement for supporting the increasing density requirements of boards, modules, and cards. Specifically, the proposed TTP design allows for die-based cooling of multiple high-power chips on dense boards, modules, and / or cards. In some embodiments, a TTP-integrated cold plate (e.g., a wet TTP) is provided, combining a single component to act as both a TTP and a cold plate. This proposal allows for die-based design, improved thermal management of chips, enhanced warpage control, and improved dense baseboard and module designs.

[0007] Another embodiment of this disclosure proposes a TTP with a separate cold plate, wherein the TTP can be disposed on one side of a heat spreader, and wherein the cold plate can be disposed on the opposite side of the heat spreader. The TTP and the cold plate can cooperate to remove heat generated by the chip located beneath the TTP and the cold plate.

[0008] This disclosure includes the following capabilities: using copper tubing loops to guide coolant through the TTP for use as a cold plate or wet TTP; using high conductivity spreaders to contact high-power chips; conducting heat generated by the high-power chips back to the TTP via the diffuser and spacer pads; and using an integrated TTP cold plate (e.g., a wet TTP) as a manifold to control the coolant flow rate. Wet TTPs can use parallel manifolds to control the flow rate, ensuring uniform coolant distribution across all high-power chips, thereby enhancing cooling performance.

[0009] This disclosure offers several advantages. First, it allows for die-based correspondence to multiple high-power chips on a backplane, module, and / or PCIe device. This configuration achieves effective cooling without compromising board density. Second, the TTP and / or cooling plate according to embodiments of this disclosure can utilize a simple design with a single loop, thereby reducing the risk of leakage. Third, embodiments of this disclosure offer superior thermal performance compared to board-based designs. Fourth, embodiments of this disclosure reduce the number of required mounting holes and achieve efficient thermal management by using a common cold plate with copper diffusers for each chip. The requirement for mounting holes on the backplane is significantly reduced because only holes are needed in the backplane to mount the "TTP integrated cold plate." The copper diffuser for each chip is directly attached to the TTP and conducts heat to the common cold plate, ensuring efficient thermal management. Fifth, embodiments of this disclosure enable the use of a single cold plate die-based correspondence to the chip, reducing voltage drop and improving reliability and overall design robustness. Sixth, the TTP integrated cold plate also acts as a manifold, thereby precisely controlling the coolant flow rate. This helps optimize the cooling performance of multiple high-power chips. The bends and size of the copper tubes can be adjusted to ensure that the coolant flow is neither too high nor too low, thereby optimizing cooling efficiency. For example, under constant pressure drop conditions of 4.8 psi, the flow rate with different copper tube inner diameters can be described as follows: (3mm inner diameter - 0.55 LPM; 4mm inner diameter - 0.86 LPM; 5mm inner diameter - 1.24 LPM; 6mm inner diameter - 2.22 LPM). For a 3mm inner diameter, the flow rate with the number of bends can be described as follows: (5 bends - 1.7 LPM; 10 bends - 0.85 LPM; 15 bends - 0.55 LPM; 20 bends - 0.42 LPM).

[0010] A system according to embodiments of the present disclosure may include: a substrate supporting one or more processors; and a heat transfer plate including one or more slits for receiving and accommodating one or more processors. The heat transfer plate may be configured to transfer heat away from one or more processors.

[0011] The system according to other embodiments of this disclosure may include: a substrate; a plurality of processors mounted on the substrate; and a heat transfer plate including a plurality of slits, wherein each slit corresponds to a different processor among the plurality of processors with respect to a die. The heat transfer plate may include fluid conduits therein that deliver coolant around the plurality of processors to extract heat generated by the plurality of processors.

[0012] A system according to other embodiments of this disclosure may include a base plate supporting a plurality of computing components. A heat transfer plate having a plurality of notches is fixed to the base plate, wherein each notch corresponds to a different computing component among the plurality of computing components with respect to a wick. At least one diffuser facilitates heat transfer from the upper surface of at least one of the plurality of computing components to a heat transfer fluid conduit.

[0013] In some embodiments, a system is provided comprising: a substrate supporting one or more processors; and a heat transfer plate including one or more slits for receiving and accommodating one or more processors, wherein the heat transfer plate is configured to transfer heat away from one or more processors.

[0014] In some aspects, one or more processors include at least one of a graphics processing unit (GPU) and a central processing unit (CPU).

[0015] In some aspects, the system also includes fluid conduits that transport heat transfer fluid around one or more processors.

[0016] In some cases, fluid conduits are integrated into the heat transfer plate.

[0017] In some aspects, the system also includes highly conductive diffusers disposed on each of one or more processors to extract heat from the top surface of one or more processors, wherein each highly conductive diffuser is also thermally coupled to a heat transfer plate to remove heat from each highly conductive diffuser.

[0018] In some aspects, one or more processors include multiple processors, and the heat transfer plate is mounted to the substrate using a number of fasteners, the number of which is twice the number of processors in the multiple processors.

[0019] In some aspects, each highly conductive diffuser is secured to the heat transfer plate using one or more additional fasteners.

[0020] In some cases, one or more additional fasteners include spring screws.

[0021] In some cases, fluid conduits are embedded in the heat transfer plate so that the heat transfer plate can operate as a wet heat transfer plate.

[0022] In some aspects, a heat transfer plate is mounted on the same surface of a substrate as one or more processors, and the heat transfer plate is thermally coupled to the surface of each of the processors in one or more processors.

[0023] In some aspects, the system also includes one or more gap pads.

[0024] In some respects, each of the one or more cuts corresponds to a corresponding processor in one or more processors, with the die as the reference.

[0025] In some aspects, the heat transfer plate includes a fluid conduit incorporated therein, which operates as a manifold to control the flow rate of the fluid passing through it.

[0026] In some aspects, manifolds include parallel manifolds that ensure the flow of fluid is distributed substantially uniformly across one or more processors.

[0027] In some respects, the substrate is part of the base plate.

[0028] In some respects, the substrate is part of the PCIe card.

[0029] In some embodiments, a system is provided comprising: a substrate; a plurality of processors mounted on the substrate; and a heat transfer plate including a plurality of slits, wherein each slit corresponds to a different processor among the plurality of processors with respect to a die, wherein the heat transfer plate includes fluid conduits therein for delivering coolant around the plurality of processors to extract heat generated by the plurality of processors.

[0030] In some aspects, fluid conduits include an inlet for receiving coolant and an outlet for discharging coolant.

[0031] In some aspects, the system also includes multiple highly conductive diffusers, each of which is positioned above a corresponding processor in a plurality of processors, and each highly conductive diffuser is fastened to the top surface of a heat transfer plate to thermally couple the top surface of the processor to the heat transfer plate.

[0032] In some aspects, the system also includes: a plurality of gap pads, each gap pad being disposed between a high-conductivity diffuser and a heat transfer plate in a plurality of high-conductivity diffusers; and a thermal interface material (TIM) disposed between a top surface of the processor and a bottom surface of a corresponding high-conductivity diffuser, wherein the TIM includes at least one of the following: thermal grease, thermal tape, thermal pad, gel, adhesive, gap filler, and phase change material.

[0033] In some embodiments, a system is provided comprising: a base plate supporting a plurality of computing components; a heat transfer plate fixed to the base plate, the heat transfer plate including a plurality of cutouts, wherein each cutout corresponds to a different computing component among the plurality of computing components with respect to a ferrule; and at least one diffuser facilitating heat transfer from an upper surface of at least one of the plurality of computing components to a heat transfer fluid conduit.

[0034] Any aspect combined with any or more other aspects.

[0035] Any one or more of the features disclosed in this document.

[0036] As is substantially disclosed herein, any one or more of the features.

[0037] Any one or more features substantially disclosed herein combined with any one or more other features substantially disclosed herein.

[0038] Any aspect / feature / implementation can be combined with any one or more other aspects / features / implementations.

[0039] Use any one or more aspects or features as disclosed herein.

[0040] It should be understood that any feature described herein may be claimed in combination with one or more other features as described herein, regardless of whether such features are derived from the same implementation.

[0041] Details of one or more aspects of this disclosure are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the technology described in this disclosure will become apparent from the specification, drawings, and claims.

[0042] Many additional features and advantages of this disclosure will become apparent to those skilled in the art when considering the following description of the implementations provided.

[0043] Additional features and advantages are described herein and will become apparent from the following description and figures. Attached Figure Description

[0044] The present disclosure has been described in general terms. Now, reference is made to the accompanying drawings, which are not necessarily drawn to scale: Figure 1A This is a block diagram of a system utilizing a heat transfer plate according to at least some embodiments of the present disclosure; Figure 1B This is a side view of a system utilizing a heat transfer plate according to at least some embodiments of the present disclosure; Figure 2A This is a top front perspective view of a system utilizing a heat transfer plate according to at least some embodiments of the present disclosure; Figure 2B yes Figure 2A Bottom front perspective view of the system; Figure 3 yes Figure 2A Top rear perspective view of the system; Figure 4 yes Figure 2A The top front perspective view of the system, in which some components have been removed to expose other underlying components; Figure 5 yes Figure 2A A top-down plan view of the system; Figure 6 yes Figure 2A A top-down plan view of the system, in which some components have been removed to expose other underlying components; Figure 7 Based on at least some embodiments of this disclosure Figure 2A Top perspective view of the heat transfer plate of the system; Figure 8 Based on at least some embodiments of this disclosure Figure 2A Bottom perspective view of the system's heat spreader; Figure 9 This is a block diagram of another system utilizing a heat transfer plate according to at least some embodiments of the present disclosure; Figure 10 This is a block diagram illustrating a computer system according to at least some embodiments of the present disclosure; Figure 11 This is a block diagram illustrating further details of a data center and its components according to at least some embodiments of the present disclosure; Figure 12 This is a block diagram of a system according to at least some embodiments of the present disclosure; Figure 13 A data center according to at least some embodiments of the present disclosure is depicted; Figure 14 A liquid-cooled server rack according to at least some embodiments of the present disclosure is depicted; Figures 15A-15D A block diagram is depicting aspects of a means for liquid cooling of racks to support neural networks at the data center level, according to at least some embodiments of the present disclosure. Figure 16 Block diagrams of systems according to at least some embodiments of the present disclosure are depicted; and Figure 17 A block diagram depicts a computing environment according to at least some embodiments of the present disclosure.

[0045] The same figures and labels in the various figures may indicate the same elements. Detailed Implementation

[0046] The following description provides only examples and is not intended to limit the scope, applicability, or configuration of the claims. Rather, it is intended to provide a credible description for those skilled in the art to implement the described embodiments. It should be understood that various changes may be made to the function and arrangement of the elements without departing from the spirit and scope of the appended claims.

[0047] As used in this article, the phrases “at least one,” “one or more,” “or,” and “and / or” are open-ended expressions that function as both conjunctions and adversative conjunctions in action. For example, each of the expressions “at least one of A, B, and C,” “at least one of A, B, or C,” “one or more of A, B, and C,” “one or more of A, B, or C,” “A, B, and / or C,” and “A, B, or C” implies: A alone, B alone, C alone, A and B together, A and C together, B and C together, or A, B, and C together.

[0048] Various aspects of this disclosure are described herein with reference to the accompanying drawings, which are schematic illustrations of idealized configurations.

[0049] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It should also be understood that terms such as those defined in common dictionaries shall be interpreted as having the same meaning as they have in the context of the relevant art and this disclosure.

[0050] Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” used herein are intended to include the plural forms as well. It should also be understood that the terms “comprise,” “comprises,” and / or “comprising,” when used in this specification, specify the presence of the stated feature, integer, step, operation, element, and / or component, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. The term “and / or” includes any and all combinations of one or more of the associated listed items.

[0051] The present disclosure is described more fully below with reference to the accompanying drawings, which illustrate some, but not all, embodiments of the disclosure. In fact, the present disclosure may be implemented in many different forms and should not be construed as being limited to the embodiments set forth herein.

[0052] The same numbers throughout refer to the same elements. As used herein, terms such as “front,” “back,” “top,” etc., are used in the examples provided below to describe the location of certain components or portions of components in an installed and operable configuration. As used herein, the term “module” encompasses hardware, software, and / or firmware configured to perform one or more specific functions, including but not limited to the conversion and transmission of electrical and optical signals. According to this disclosure, it will be apparent to those skilled in the art that the term “substantially” indicates that the referenced element or associated description is accurate within applicable engineering tolerances.

[0053] The embodiments of this disclosure are intended to be deployed in data center environments. While embodiments have been described in conjunction with certain examples of data center environments, it should be understood that the embodiments of this disclosure are not limited thereto. In fact, the embodiments of this disclosure can be deployed in any number of environments, including data center environments or any other suitable environment that requires managing the thermal load generated by computing components.

[0054] Now, for reference Figures 1A to 17 A heat transfer plate for die-based cooling according to embodiments of the present disclosure is shown and described, as well as illustrative data center environments and components for a full understanding of the present disclosure.

[0055] Figure 1A and Figure 1B A system 100 is depicted that utilizes a heat transfer plate 112 to facilitate cooling of multiple computing components mounted to a substrate or base plate 104. In some embodiments, the computing components may include one or more processors or chips 108. The heat transfer plate 112 may include one or more cutouts 106 for accommodating and receiving the chips 108, wherein the heat transfer plate 112 is configured to transfer heat away from the chips 108.

[0056] The substrate or baseplate 104 may be, for example, a board, module, and / or a PCIe card baseplate. The substrate 104 may be, for example, a printed circuit board (PCB). The substrate 104 supports multiple computing components, including multiple chips or processors 108. One or more of the chips or processors 108 may be, for example, a graphics processing unit (GPU), a GPU package, a central processing unit (CPU), a CPU package, a quantum processing unit (QPU), a QPU package, a parallel processing unit (PPU), a PPU package, an application-specific integrated circuit (ASIC), and / or any other processing unit or similar silicon package that generates heat during use. (Throughout this disclosure, it should be understood that references to CPU, CPU package, GPU, and GPU package also apply to other types of processing units and processing unit packages, including any other types of processing units and processing unit packages described herein.) The multiple computing components supported by the substrate 104 may also include multiple low-power components 110, such as... Figure 1B As shown. Within the meaning of this disclosure, "low-power component" includes any component that draws relatively less power (and therefore generates relatively less heat) than a CPU, GPU, or other processing unit or processing unit package on a motherboard, motherboard, blade, node, or other computing device. Thus, for example, a low-power component can be or includes random access (or similar) memory cells, chipset or platform controller hubs, high-speed interconnects, backplane management controllers, memory controllers, voltage regulators, and / or power delivery circuitry. In some embodiments, a low-power component can be or includes any component that draws less than about 100W of power during operation.

[0057] The heat transfer plate (TTP) 112 can be mounted to the substrate via multiple mounting tabs, and as... Figure 1A As shown, it includes a single integral piece having a die-based cutout 106 for accommodating and receiving chip 108. The TTP 112 can be mounted to directly contact multiple low-power components 110, although in some embodiments, a thermal interface material (TIM) layer or spacer pad can be used to facilitate thermal coupling between the TTP 112 and one or more low-power components 110. When using a thermal interface material, the TIM can be grease, thermally conductive tape, thermally conductive pad, gel, adhesive, gap filler, or phase change material.

[0058] TTP 112 can be made of copper, aluminum, copper alloys, aluminum alloys, other metals, other metal alloys, and / or other thermally conductive materials. The thickness of TTP 112 can be selected to obtain sufficient structural rigidity to ensure the desired level of heat transfer and / or minimize the required thickness of any thermal interface material (TIM) and / or spacer pads used to thermally couple chip 108 to TTP 112 via highly conductive diffuser 120, as described below. In some embodiments, different segments of TTP 112 can have different thicknesses, for example, to facilitate thermal coupling of TTP 112 to computing components of different heights. TTP 112 can be formed into its final shape using one or more of machining, additive manufacturing, powder metallurgy, extrusion, casting, and forging.

[0059] As used in this article, the term “thermal coupling” refers to two or more objects connected in a manner that facilitates heat transfer between the two objects, such that from a heat transfer perspective, the two or more objects act as a single entity (or essentially act as a single entity).

[0060] System 100 also includes a plurality of high-conductivity spreaders 120. The high-conductivity spreaders 120 are shaped to have a bottom portion extending into each notch in the TTP 112 for thermal coupling with the corresponding chip 108, and a top portion extending outward for thermal coupling with the TTP 112. Similar to the TTP 112, the high-conductivity spreaders 120 can be made of copper, aluminum, copper alloys, aluminum alloys, other metals, other metal alloys, and / or other high thermal conductivity materials. The thickness of the high-conductivity spreaders 120 (including the thickness of the bottom and top portions of each high-conductivity spreader 120) can be selected to obtain sufficient structural rigidity to ensure the desired level of heat transfer and / or minimize the required thickness of any thermal interface material (TIM) and / or spacer pads used to thermally couple the high-conductivity spreaders 120 to the chip 108 and / or TTP 112.

[0061] Each high-conductivity diffuser 120 is thermally coupled to the chip 108 using a thermal interface material (TIM) 124, which helps ensure efficient heat transfer from the chip 108 and the high-conductivity diffuser 120. In some embodiments, spacer pads may be used to thermally couple one or more chips of the chip 108 to the high-conductivity diffuser 120.

[0062] Each highly conductive diffuser 120 is also thermally coupled to the TTP 112 via a plurality of spacers 128, which also helps to ensure efficient heat transfer from the highly conductive diffuser 120 to the TTP 112. In some embodiments, a thermal interface material may be used to thermally couple one or more of the highly conductive diffusers 120 to the TTP 112. While thermal interface materials can provide better thermal conductivity than spacers, spacers advantageously facilitate maintenance due to their ease of removal and replacement.

[0063] The high conductivity diffuser 120 is directly attached to the TTP 112. In system 100, spring screws 132 are used to secure each high conductivity diffuser 120 to the TTP 112. The spring screws 132 advantageously secure the high conductivity diffuser 120 to the TTP 112 while allowing the high conductivity diffuser 120 to “float” above the chip 108, thereby avoiding damage to the chip 108 and promoting better contact between each high conductivity diffuser 120 and the spacer pad 128, TIM 124, and / or the chip 108 beneath the high conductivity diffuser 120. However, in other embodiments, other attachment modes may be utilized, including other mechanical fasteners (e.g., bolts, screws, locking tabs), adhesives, and / or bonding methods.

[0064] The TTP 112 of system 100 can be configured to operate as a cooling plate, wherein conduits pass through the cooling plate (e.g., are embedded in the cooling plate), and a heat transfer fluid (also referred to as coolant throughout this disclosure) is pumped through the conduits to facilitate heat extraction from the TTP 112. Thus, system 100 also includes a heat transfer fluid inlet 136 and a heat transfer fluid outlet 140. The size and shape of the conduits (including the cross-sectional shape and geometry of the conduits) can be selected to achieve a desired level of heat transfer from the TTP 112 (e.g., a level that ensures heat does not degrade the operation of chip 108 and / or low-power components 110). Although system 100 includes only one inlet 136 and one outlet 140, systems according to other embodiments of this disclosure may utilize multiple inlets 136 and / or outlets 140 (connecting multiple conduits extending through the TTP 112) to facilitate adequate cooling of each chip in chip 108 being cooled by the TTP 112. More specifically, multiple conduits can be used to ensure that the coolant flow is evenly distributed across all the high-power chips 108.

[0065] As computing components are increasingly densely packaged on substrates or baseplates of boards, modules, and PCIe cards, the space available for mounting cooling components to such substrates or baseplates becomes increasingly limited. This is particularly problematic for cooling high-power components such as chip 108, as there may not be sufficient space on the substrate (such as substrate 104) to mount a die-based cold plate, wet TTP, or other cooling structure to the substrate or baseplate for each high-power component. The TTP 112 advantageously solves this problem by mounting to the substrate using a relatively small number of mounting tabs, while providing additional surface to which the die-based high-conductivity diffuser 120 can be secured. The number of mounting tabs used to secure the TTP 112 to the substrate 104 can, for example, be equal to or less than twice the number of chips 108 surrounded by the TTP 112, or in other words, equal to or less than twice the number of high-conductivity diffusers 120 supported (or capable of being supported by) the TTP 112. Figure 1B As shown, each high conductivity diffuser in the high conductivity diffuser 120 corresponds to a corresponding chip or processor 108 with the die as a reference, thereby ensuring that each chip 108 receives proper cooling.

[0066] During operation of system 100, heat generated by chip 108 is transferred to high-conductivity diffuser 120 via TIM 124. Some additional heat transfer may occur directly from chip 108 to TTP 112. The heat received by high-conductivity diffuser 120 from chip 108 is then transferred to TTP 112 via spacer pad 128. Using die-based high-conductivity diffuser 120 to transfer heat to TTP 112 (rather than attempting to utilize a solid heat transfer plate (or any other cold plate) simply mounted to the substrate) ensures efficient heat transfer from chip 108 regardless of its height or orientation. Because individual high-conductivity diffusers 120 can be removed to access individual chip 108 without removing the entire TTP 112, TTP 112 also improves the maintainability of chip 108, meaning that other chips 108 can continue to operate during the maintenance of individual chip 108.

[0067] Figures 2A to 8 Various aspects of a heat transfer plate system 200 according to embodiments of the present disclosure are depicted, and although reference may be made to specific drawings in describing particular aspects of the heat transfer plate system 200, the same or similar components may be used. Figures 2A-8 It is depicted in other accompanying figures.

[0068] The heat transfer plate system 200 includes a heat transfer plate 212. The TTP 212 may be the same as or similar to the TTP 112. More specifically, the TTP 212 may be made of copper, copper alloys, aluminum, aluminum alloys, any other metal or metal alloy, and / or any other thermally conductive material. The TTP 212 may be formed into its final shape using one or more of machining, additive manufacturing, powder metallurgy, extrusion, casting, and forging.

[0069] TTP 212 includes multiple die-based cutouts 206 (e.g., Figure 7 As shown), each cutout is positioned to provide a passage to the chip or processor 208 below the TTP 212. The chip or processor 208 may be the same as or similar to the chip or processor 108. Thus, for example, one or more chips or processors of the chip or processor 208 may be or include a graphics processing unit (GPU), a GPU package, a central processing unit (CPU), a CPU package, a quantum processing unit (QPU), a QPU package, a parallel processing unit (PPU), a PPU package, an application-specific integrated circuit (ASIC), and / or any other processing unit or similar silicon package that generates heat during use.

[0070] Multiple die-based high thermal conductivity heat diffusers 220 are mounted to the TTP 212 using multiple spring screws 232. The heat diffusers 220, which may be the same as or similar to the high thermal conductivity diffuser 120, are thermally coupled to the chip or processor 208 and the TTP 212, and are used to transfer heat from the chip 208 to the TTP 212. The TTP system 200 uses four spring screws 232 to secure each heat diffuser 220 to the TTP 212, but in other embodiments, more or fewer spring screws 232 may be used. Additionally, in other embodiments, fasteners other than spring screws or attachment modes other than fasteners may be used to secure the heat diffusers 220 to the TTP 212. In some embodiments, for example, the heat diffuser 220 may be fixed to the TTP 212 (directly or indirectly through thermal interface material and / or one or more spacer pads) using an adhesive, or the heat diffuser 220 may simply rest on (e.g., without being fixed to) the TTP 212 (directly or indirectly through thermal interface material and / or one or more spacer pads).

[0071] Although Figure 2A-8Not shown, but a thermal interface material or spacer pad can be used to facilitate heat transfer between each thermal diffuser 220 and the corresponding chip 208. More specifically, the thermal interface material or spacer pad can be placed between the thermal diffuser 220 and the chip 208 to ensure contact across the entire top surface of the chip 208 and the entire bottom surface of the thermal diffuser 220. When using a thermal interface material, the TIM can be any one or more of the following: grease, thermally conductive tape, thermally conductive pad, gel, adhesive, gap filler, and phase change material.

[0072] Each individual heat diffuser 220 is mounted to the TTP 212 using spring screws 232. This advantageously ensures that each heat diffuser 220 can be die-referenced to the corresponding chip 208, while multiple mounting tabs 216 are used to directly mount the larger TTP to the same substrate as the chip 208. This arrangement reduces the number of mounting locations required on the substrate, which may be densely packed with computing components, while still allowing for die-referenced placement of the heat diffusers 220. The spring screws 232 also advantageously ensure that a desired level of force—neither too large (which could damage the underlying chip 208) nor too small (which could adversely affect heat transfer from the chip 208 and / or the TTP 212 to the heat diffuser 220)—is applied to the heat diffuser 220 in the direction of the TTP 212 and the chip 208.

[0073] Each heat diffuser 220 is individually mounted to the TTP 212 using spring screws 232, which further advantageously allows for the individual removal of each heat diffuser 220, for example, to access the chip 208 for maintenance or replacement. Therefore, using the TTP system 200 to cool the chip 208 improves the maintainability of the underlying board, module, PCIe card, or other computing device that includes the chip 208.

[0074] Each mounting tab in the mounting tabs 216 is configured to receive a fastener (not shown) that secures the mounting tab 216, and thus the TTP 212, to the underlying substrate or other mounting component. The mounting tabs 216 may also be or include risers or spacers to ensure that the TTP 212 provides sufficient clearance for computing devices (including, for example, chips or processors 208) mounted to the underlying substrate when mounted to it. As noted above, the number of mounting tabs 216 on the TTP 212, and therefore the number of mounting holes and connection points required in the underlying substrate, is less than the total number of mounting holes that might be required in the substrate when mounting individual die-based cooling systems to a substrate for each chip 208. For example, the number of mounting holes and connection points for the TTP 212 of this disclosure may include approximately 12, while the number of mounting holes and connection points for a non-die-based TTP as described herein may be up to 32. Therefore, since the TTP 212 can be mounted to a relatively small number of locations on the substrate using mounting tabs 216, while the die-based heat diffuser 220 can be mounted to as many mounting locations on the TTP as needed, the TTP 212 is particularly useful for densely packaged substrates. For example, in the TTP system 200, the total number of mounting tabs is 10, while the total number of spring screws 212 used to mount the die-based heat diffuser 220 to the TTP 212 is 32. In some embodiments, the number of mounting tabs 216 on the TTP 212 may be equal to or less than twice the number of chips 208 cooled by the TTP 212 (or in other words, equal to or less than twice the number of heat diffusers 220 mounted to the TTP 212). In some embodiments, the number of mounting tabs 216 on the TTP 212 may be equal to the number of chips 208 cooled by the TTP 212 (or in other words, equal to the number of heat diffusers 220 mounted on the TTP 212).

[0075] The thickness of TTP 212 can be selected to obtain sufficient structural rigidity to ensure the desired level of heat transfer, and / or to minimize the required thickness of any thermal interface material (TIM) and / or spacer pads used to thermally couple the chip 208 of the TTP system 200 to the TTP 212 via the thermal diffuser 220. In some embodiments, different segments of the TTP 212 may have different thicknesses, for example, to facilitate thermal coupling of the TTP 212 to computing components of different heights.

[0076] For example, such as Figure 2BAs seen, the TTP 212 can be configured to be directly cooled (e.g., without the use of a heat diffuser 220) and mounted on an additional computing component 210 on the same substrate as the chip 208 and the TTP 212. The computing component 210 can be or includes low-power components, as defined above. Such components do not require the cooling levels required by high-power components such as the chip 208, and therefore can be adequately cooled simply by being placed in contact with the thermally conductive TTP 212. The bottom surface of the TTP 212 may accordingly include a plurality of downwardly extending protrusions 214, each protrusion 214 being positioned and sized to contact the upper surface of the low-power component for thermal coupling to the low-power component for cooling the low-power component. In some embodiments, a spacer pad or thermal interface material (including any TIM described herein) may be used between the low-power component and the TTP 212 to facilitate efficient thermal coupling between them.

[0077] In addition, Figure 2B as well as Figures 6-7 Mounting holes 234 are visible in TTP 212. These mounting holes 234 are configured to receive spring screws 232 for mounting the heat diffuser 220 to TTP 212. When using fasteners other than spring screws 232 to secure the heat diffuser 220 to TTP 212, the mounting holes 234 may be appropriately modified (or omitted) depending on the type of fastener, attachment method, or mechanism used.

[0078] TTP 212 includes a heat transfer fluid inlet 236 and a heat transfer fluid outlet 240. The heat transfer fluid inlet 236 and the heat transfer fluid outlet 240 may be the same as or similar to the heat transfer fluid inlet 136 and the heat transfer fluid outlet 140 of the heat transfer plate system 100, respectively. Each of the heat transfer fluid inlet 236 and the heat transfer fluid outlet 240 includes a heat transfer fluid interface 248 for connecting a heat transfer fluid circulation system to the TTP system 200. More specifically, heat transfer fluid is received at inlet 236 through one interface 248 from a heat transfer fluid source (e.g., a heat transfer fluid reservoir and / or a cooling system circulating heat transfer fluid among multiple computing systems) and discharged from outlet 240 to the heat transfer fluid source through the other interface 248. The heat transfer fluid interface 248 may include, for example, a locking or other attachment mechanism for securing an external heat transfer fluid conduit to the heat transfer fluid interface 248.

[0079] The heat transfer fluid entering the heat transfer fluid inlet 236 is guided to the fluid conduit 244 (the parts of which are in Figure 2A-2B and Figure 4-5 As can be seen, and its whole is in Figure 6(As shown in the diagram). Fluid conduit 244 then guides the heat transfer fluid through TTP 212, surrounding each chip in chip 208 cooled by TTP 212. Conduit 244 is a copper tube, but in other embodiments it can be a tube or tube ring made of any thermally conductive material, including copper, copper alloys, aluminum, aluminum alloys, any other metal or metal alloy, and / or non-metallic but thermally conductive materials. Specific geometries of conduit 244 (including the number and radius of bends in conduit 244) and the cross-sectional area of ​​conduit 244 can be selected to maximize cooling efficiency and / or achieve a specific desired cooling level. In other words, the bends and size of conduit 244 can be selected to ensure that the coolant flow is neither too high nor too low, thereby optimizing cooling efficiency.

[0080] Due to the geometry and cross-sectional area of ​​conduit 244, conduit 244 (and more broadly, TTP system 200) acts as a manifold to control the coolant flow rate. In some embodiments, a TTP such as TTP 212 may include multiple conduits and / or parallel manifolds for controlling the heat transfer fluid flow rate, thereby ensuring uniform distribution of coolant flow across all high-power chips 208, thus enhancing cooling performance.

[0081] Figure 4 A TTP system 200 is shown, in which a single highly conductive heat diffuser 220 is removed from a TTP 212. With the heat diffuser removed, a portion of a conduit 244 is visible, which bends around a cutout 206 in the TTP 212, providing access to the underlying chip 208. Additionally, two spacer pads 228 are located on opposite sides of the cutout 206 above the conduit 244; these facilitate heat transfer from the heat diffuser 220 (when it is installed) to the conduit 244 (and into the heat transfer fluid passing through the conduit 244). In some embodiments, a thermal interface material is used instead of the spacer pads. However, because spacer pads are easier to remove and replace, using spacer pads instead of thermal interface materials (e.g., phase change materials) simplifies the maintenance process. More specifically, when maintenance of the chip 208 of the TTP system 200 is required, the corresponding heat diffuser 220 can be easily removed by loosening the spring screw 232, after which the spacer pads 228 can also be removed. In short, the gap pad 228 reduces the complexity of maintenance, and therefore also reduces the time required for maintenance.

[0082] Figure 5 and Figure 6 A top view of the TTP system 200 is shown, although in Figure 6All heat diffusers 220 (and their corresponding spring screws 232 and spacers 228) have been removed to provide a clear view of the components beneath. In one or both of these views, both the fastener holes 218 in the mounting tabs 216 and the mounting holes 234 for the spring screws 232 are visible. The size of these holes can be selected based on the desired fasteners for mounting the TTP 212 to the underlying substrate and for mounting the heat diffusers 220 to the TTP 212, respectively.

[0083] like Figure 6 As can be seen, the conduit 244 bends around each die-based cutout 206, thereby providing cooling to three sides of the TTP 212 surrounding each cutout 206. Between each cutout 206, the conduit 244 extends toward the opposite side of the TTP 212, thereby providing cooling to the TTP 212 away from the cutout 206 in areas where the TTP 212 may come into thermal contact with, for example, low-power components on the underlying board, module, PCIe card, or other computing device.

[0084] Although conduit 244 is shown as having a specific path extending from heat transfer fluid inlet 236 to heat transfer fluid outlet 240, in other embodiments of this disclosure, conduit 244 may follow a different path. For example, the path of conduit 244 may have bends with large radii of curvature, and / or may have more or fewer bends. In some embodiments, conduit 244 may extend from heat transfer fluid inlet 236 to heat transfer fluid outlet 240 in a straight or substantially straight line. Furthermore, in some embodiments, conduit 244 may extend along the length of TTP 212 (e.g., from heat transfer fluid inlet 236 to near heat transfer fluid outlet 240), then turn back to near heat transfer fluid inlet 236, and then turn back again to heat transfer fluid outlet 240.

[0085] In other embodiments, the heat transfer fluid inlet 236 may be positioned adjacent to the heat transfer fluid outlet 240, and the conduit 244 may transport heat transfer fluid from one side of the TTP 212 through the TTP 212 and return to the same side of the TTP 212. In various embodiments of this disclosure, one or more heat transfer fluid inlets 236 and / or heat transfer fluid outlets 240 may be located on either side of the TTP 212. For example, a first pair including heat transfer fluid inlets and outlets may be located on a first half of the TTP 212, with a first conduit extending therebetween (in the first half of the TTP 212), and a second pair including heat transfer fluid inlets and outlets may be located on the other half of the TTP 212, with a second conduit extending therebetween (in the second half of the TTP 212). Because the heat transfer fluid passes through shorter conduits, its temperature is lower as it passes through the last portion of each conduit, thus such embodiments can advantageously achieve improved cooling.

[0086] Furthermore, in some embodiments, a TTP such as TTP 212 may include two or more conduits 244. For example, in cases requiring particularly high cooling levels, two or more conduits such as conduit 244 may each deliver heat transfer fluid through the TTP, thereby increasing the total volume of heat transfer fluid through the TTP while reducing the area of ​​the TTP cooled by each conduit. In such embodiments, the TTP system may be provided with separate heat transfer fluid inlets and outlets for each conduit, or a single heat transfer fluid inlet and outlet may be configured to deliver heat transfer fluid to and receive heat transfer fluid from multiple conduits, respectively.

[0087] Although the conduit 244 of the TTP system 200 is shown to have a specific size, different sizes of conduits may be used in various embodiments of this disclosure. For example, the conduit 244 may have a larger or smaller width, height, and / or length. The conduit 244 may have a square, rectangular, circular, elliptical, triangular, or other cross-sectional shape.

[0088] The size, cross-sectional shape, and geometry of the conduit 244 can be advantageously selected to provide a desired level of heat transfer fluid flow through the TTP 212 and / or a desired level of cooling to the TTP 212. In other words, the conduit 244 (and more generally, the TTP system 200) can function as a manifold, thereby allowing precise control of the coolant flow rate and allowing for optimization of cooling efficiency.

[0089] Figure 7Only TTP 212 of TTP system 200 is shown. As can be seen in this view, TTP 212 includes a conduit cutout 246 in which a conduit 244 is placed; and a die-based chip cutout 206 through which a heat diffuser 220 extends to achieve thermal contact with a corresponding chip 208 located below each cutout. A plurality of mounting tabs 216 facilitate mounting TTP 212 to an underlying substrate, and holes 218 extending through TTP 212 and each mounting tab 216 are configured to receive appropriate fasteners for securing TTP 212 to the underlying substrate. In this embodiment, the holes 218 located on the body of TTP 212 itself (rather than on the mounting tabs 216) are countersunk holes; however, in other embodiments, such holes 218 may alternatively be countersunk holes, or may not be countersunk holes or not countersunk holes.

[0090] The overall geometry of the TTP 212 can be selected or defined based on the specific application in which it will be used. Thus, for example, in some embodiments, the top surface of the TTP 212 can be substantially planar (although it may have various cutouts and holes to accommodate, for example, a heat diffuser 220, a conduit 244, mounting screws 232, and fasteners for mounting the TTP 212 to the underlying substrate). In other embodiments (including...) Figure 2A-8 In the depicted embodiment, TTP 212 may include a stepped surface, wherein the front portion of TTP 212 has a substantially planar top surface at a first level, and the rear portion of TTP 212 has a substantially planar top surface at a second level different from the first level.

[0091] like Figure 8 As shown, the heat diffuser 220 includes a bottom portion 854 and a top portion 856. The bottom portion 854 is sized to fit within a die-based cutout 206 on the TTP 212, while the top portion is sized to extend significantly beyond the sides of the cutout 206 and provide space for the heat diffuser 220 to extend above the conduit 244 passing through the TTP 212 and also above the TTP 212 itself, wherein a mounting hole 858 can be configured to receive a spring screw 232 for securing the heat diffuser 220 to the TTP 212. The bottom surfaces 852 of the top portion 856 and the bottom surface of the bottom portion 854 of the heat diffuser 220 are substantially planar to facilitate robust contact with the TTP 212 and the chip 208, respectively. When thermal interface material and / or spacer pad 228 are used to ensure thermal coupling between thermal diffuser 220 and TTP 208, conduit 244 and / or chip 208, the substantially planar bottom surface of thermal diffuser 220 promotes robust contact between thermal diffuser 220 and thermal interface material and / or spacer pad 228.

[0092] Figure 2A-8 The TTP system 200 shown includes several features. System 200 guides coolant (i.e., heat transfer fluid) through TTP 212 in conduit 244, thereby allowing TTP 212 to operate as a wet TTP. A heat diffuser 220 with high thermal conductivity contacts the corresponding high-power chip 208 directly or indirectly (e.g., through a thermal interface material or spacer). The heat diffuser 220 conducts heat generated by the high-power chip 208 directly or through the spacer back into the TTP and into conduit 244. TTP system 200 acts as a manifold to control the coolant flow rate. In some embodiments, TTP system 200 includes parallel conduits and / or inlet / outlet manifolds for controlling the coolant flow rate and ensuring uniform distribution of coolant flow across all high-power chips 208, thereby enhancing cooling performance.

[0093] Now, turn to Figure 9 According to another embodiment of the present disclosure, a heat transfer plate system 900 includes a substrate or base plate 904 on which a plurality of chips or processors 908 are mounted. The substrate may be the same as or similar to the substrate or base plate 104 described above, and each of the plurality of chips or processors 908 may be the same as or similar to the chips or processors 108 and / or 208 described above. A heat transfer plate 912 is also mounted to the substrate 904 via a plurality of mounting tabs or lugs 916. Except as described below, the TTP 912 may be the same as or similar to the TTP 112 and / or TTP 212 described above. The mounting tabs or lugs 916 may be the same as or similar to the mounting tabs or lugs 116 and / or 216 described above.

[0094] Multiple die-based highly conductive heat diffusers 920 are secured to the TTP 912 using multiple spring screws 932. Each heat diffuser 920 may be identical or similar to the highly conductive diffuser 120 and / or heat diffuser 220 described above. More specifically, each heat diffuser 920 has a bottom portion extending through a die-based cutout in the TTP 912 and thermally coupled to one of the multiple chips 908 via a spacer pad 924. In some embodiments, a thermal interface material may be used instead of the spacer pad 924 to ensure thermal coupling between the heat diffuser 920 and the corresponding chip 908. Additionally, each heat diffuser 920 has a top portion extending across a portion of the top surface of the TTP 912 for thermal coupling to the TTP 912 (and to provide space for the heat diffuser 920 to be secured to the TTP 912, for example, using spring screws 932).

[0095] Each spring screw 932 may be the same as or similar to the spring screw 132 and / or spring screw 232 described above.

[0096] Although the TTP system 900 includes a heat diffuser 920 thermally coupled to the corresponding chip 908 via spacer pads 924 and in direct contact with the TTP 912, different arrangements are possible in other embodiments. For example, in other embodiments, each heat diffuser 920 may be in direct contact with the corresponding chip 908. Additionally or alternatively, each heat diffuser 920 may be thermally coupled to the surrounding portion of the TTP 912 via a thermal interface material and / or via one or more spacer pads.

[0097] The difference between TTP 912 and TTP 112 and TTP 212 is that TTP 912 does not include catheter incision 246 (e.g., as shown in the image). Figure 7 As depicted, TTP 912 also does not include coolant conduits (such as conduit 244) extending through it. Instead, a cold plate 948 located above the highly conductive heat diffuser 920 facilitates the removal of heat generated by the chip or processor 908. More specifically, the cold plate 948 is thermally coupled to the chip or processor 908 via the heat diffuser 920 and spacer pad 928 and / or via TTP 912, heat diffuser 920 and spacer pad 928.

[0098] exist Figure 9 In this embodiment, the TTP 912 supports a die-based heat diffuser 920, while advantageously requiring a relatively small number of mounting tabs 916 to mount it to a substrate 904, which can be tightly packaged with chips 908, other high-power components, and / or low-power components. The number of mounting tabs 916 used to secure the TTP 912 to the substrate 904 can, for example, be less than or equal to twice the number of chips 908 cooled by the TTP 912, and / or less than or equal to twice the number of heat diffusers 920 to which the TTP 912 is configured to receive.

[0099] TTP 912 also advantageously absorbs some heat directly from chip 908 and / or from other computing components (e.g., low-power components) mounted to substrate 904, and facilitates the transfer of this heat through heat diffuser 920 and spacer pad 928 located above heat diffuser 920 and into a separate cold plate 948 located above TTP 912, heat diffuser 920 and spacer pad 928 but thermally coupled to TTP 912, heat diffuser 920 and spacer pad 928.

[0100] The cold plate 948 may include one or more conduits therethrough for conveying heat transfer fluid through the cold plate 948. The path of the one or more conduits through the cold plate 948 may be designed to facilitate a desired level of heat transfer from the spacer 928 to the heat transfer fluid flowing through the cold plate 948. This heat transfer fluid may be received into the cold plate 948 via a heat transfer fluid inlet 936 and may exit the cold plate 948 via a heat transfer fluid outlet 940. The heat transfer fluid may be received from and / or discharged into a heat transfer fluid reservoir. In some embodiments, the cold plate 948 may be the same as or similar to any cold plate now known or developed in the future.

[0101] Now, refer to Figure 10-17 Various aspects and embodiments of a data center, including those utilizing a heat transfer plate as described above, and its components, are described. Figure 10 A computer system 1000 according to at least one embodiment is illustrated. In at least one embodiment, the systems and devices described throughout this disclosure can be implemented in conjunction with a computer system such as computer system 1000.

[0102] In at least one embodiment, the computer system 1000 includes, but is not limited to, at least one central processing unit (“CPU”) 1002, which is connected to a communication bus 1010 implemented using any suitable protocol, such as PCI (“Peripheral Component Interconnect”), PCI-Express, AGP (“Accelerated Graphics Port”), HyperTransport, or one or more other buses or point-to-point communication protocols. In at least one embodiment, the computer system 1000 includes, but is not limited to, main memory 1004. Control logic (e.g., implemented in hardware, software, or a combination thereof) and data are stored in main memory 1004, which may take the form of random access memory (“RAM”). In at least one embodiment, a network interface subsystem (“network interface”) 1022 provides an interface to other computing devices and networks for receiving data from and transferring data from the computer system 1000 to other systems.

[0103] In at least one embodiment, the computer system 1000 includes, but is not limited to, an input device 1008, a parallel processing system 1012, and a display device 1006, which may be implemented using conventional cathode ray tube (“CRT”), liquid crystal display (“LCD”), light-emitting diode (“LED”), plasma display, or other suitable display technologies. In at least one embodiment, user input is received from the input device 1008, such as a keyboard, mouse, touchpad, microphone, etc. In at least one embodiment, each of the foregoing modules may reside on a single semiconductor platform to form the processing system.

[0104] In at least one embodiment, a computer program in the form of machine-readable executable code or computer control logic algorithms is stored in main memory 1004 and / or secondary storage devices. If executed by one or more processors, the computer program enables system 1000 to perform various functions according to at least one embodiment. Memory 1004, secondary storage devices, and / or any other storage devices are possible examples of computer-readable media. In at least one embodiment, secondary storage devices can refer to any suitable storage device or system, such as hard disk drives and / or removable storage drives, representing floppy disk drives, magnetic tape drives, optical disk drives, digital versatile disc (“DVD”) drives, recording devices, Universal Serial Bus (“USB”) flash memory, etc. In at least one embodiment, the architecture and / or functionality of the various figures described herein are implemented in the context of: CPU 1002; parallel processing system 1012; integrated circuits possessing at least a portion of the capabilities of both CPU 1002 and parallel processing system 1012; chipsets (e.g., groups of integrated circuits designed to operate and be marketed as units for performing related functions); and any suitable combination of one or more integrated circuits.

[0105] In at least one embodiment, the architecture and / or functionality of the various figures described herein are implemented within the context of general-purpose computer systems, circuit board systems, game console systems dedicated to entertainment purposes, special-purpose systems, etc. In at least one embodiment, computer system 1000 may take the form of a desktop computer, laptop computer, tablet computer, server, supercomputer, smartphone (e.g., wireless handheld device), personal digital assistant (“PDA”), digital camera, vehicle, head-mounted display, handheld electronic device, mobile phone device, television, workstation, game console, embedded system, and / or any other type of logic.

[0106] In at least one embodiment, the parallel processing system 1012 includes, but is not limited to, a plurality of parallel processing units (“PPUs”) 1016 and associated memory 1016. In at least one embodiment, the PPUs 1016 are connected to a host processor or other peripheral device via interconnects 1018 and switches 1020 or multiplexers. In at least one embodiment, the parallel processing system 1012 distributes computational tasks across a plurality of PPUs 1016 that can be parallelized, for example, as part of distributing computational tasks across a plurality of graphics processing units (“GPUs”) thread blocks. In at least one embodiment, memory may be shared and accessed (e.g., for read and / or write access) among some or all of the PPUs 1016, although such shared memory may cause a performance penalty compared to using local memory and registers residing in the PPUs 1016. In at least one embodiment, the operation of the PPUs 1016 is synchronized using a command such as syncthreads(), which requires all threads in the block (e.g., executing on a plurality of PPUs 1016) to reach a certain point in the code before continuing execution.

[0107] Figure 10 Each of the CPU 1002 and PPU 1014 illustrated can be cooled, for example, using the various embodiments of this disclosure.

[0108] Figure 11 Additional components of an example data center 1100 according to at least some embodiments of the present disclosure are illustrated. The data center 1100 may also include one or more modules following one or more cooling / thermal management features as described herein.

[0109] In at least one embodiment, data center 1100 includes a data center infrastructure layer 1110, a framework layer 1120, a software layer 1130, and an application layer 1140. In at least one embodiment, the infrastructure layer 1110, framework layer 1120, software layer 1130, and application layer 1140 may be provided partially or entirely via computing components on server trays located in racks within data center 1100 (or another data center). This enables the cooling system of this disclosure to efficiently and effectively direct cooling to certain computing and interconnect features. Furthermore, various aspects of data center 1100 (including the data center infrastructure layer 1110, framework layer 1120, software layer 1130, and application layer 1140) can be used to support the selection or design of middleware layers. Thus, reference... Figure 11 The discussion can be understood as applying to, for example, the hardware and software features required to enable or support cooling functions.

[0110] In at least one embodiment, such as Figure 11As shown, the data center infrastructure layer 1110 may include a resource coordinator 1112, grouped computing resources 1114, and node computing resources (“nodes CR”) 1116(1)-1116(N), where “N” represents any positive integer. In at least one embodiment, nodes CR 1116(1)-1116(N) may include, but are not limited to, any number of central processing units (CPUs) or other processors (including accelerators, field-programmable gate arrays (FPGAs), graphics processors, etc.), memory devices (such as dynamic read-only memory), storage devices (such as solid-state or disk drives), network input / output (“NW I / O”) devices, network switches, virtual machines (“VMs”), power modules, and cooling modules, etc. In at least one embodiment, one or more of nodes CR 1116(1)-1116(N) may be servers having one or more of the computing resources mentioned above.

[0111] In at least one embodiment, the grouped computing resource 1114 may include separate groups (not shown) of node CRs housed in one or more racks, or a plurality of racks (also not shown) housed in data centers at various geographical locations. The separate groups of node CRs within the grouped computing resource 1114 may include grouped computing, networking, memory, or storage resources that can be configured or allocated to support one or more workloads. In at least one embodiment, several node CRs, including CPUs or processors, may be grouped within one or more racks to provide computing resources to support one or more workloads. In at least one embodiment, the one or more racks may also include any number of power modules, cooling modules, and network switches, in any combination.

[0112] In at least one embodiment, resource coordinator 1112 may be configured or otherwise control one or more nodes CR1116(1)-1116(N) and / or grouped computing resources 1114. In at least one embodiment, resource coordinator 1112 may include a Software Design Infrastructure (SDI) management entity for data center 1100. In at least one embodiment, resource coordinator may include hardware, software, or some combination thereof.

[0113] In at least one embodiment, such as Figure 11As shown, framework layer 1120 includes a job scheduler 1122, a configuration manager 1124, a resource manager 1126, and a distributed file system 1128. In at least one embodiment, framework layer 1120 may include a framework for supporting software 1132 of software layer 1130 and / or one or more applications 1142 of application layer 1140. In at least one embodiment, software 1132 or one or more applications 1142 may respectively include web-based service software or applications, such as service software or applications provided by Amazon Web Services, Google Cloud, and Microsoft Azure. In at least one embodiment, framework layer 1120 may be, but is not limited to, a free and open-source software web application framework, such as Apache Spark. TM (Hereinafter referred to as "Spark"), which can utilize the distributed file system 1128 for large-scale data processing (such as "big data"). In at least one embodiment, the job scheduler 1122 may include Spark drivers for facilitating the scheduling of workloads supported by various layers of the data center 1100. In at least one embodiment, the configuration manager 1124 is capable of configuring different layers, such as the software layer 1130 and the framework layer 1120, which includes Spark and the distributed file system 1128 for supporting large-scale data processing. In at least one embodiment, the resource manager 1126 is capable of managing clustered or grouped computing resources mapped to or allocated to support the distributed file system 1128 and the job scheduler 1122. In at least one embodiment, the clustered or grouped computing resources may include grouped computing resources 1114 at the data center infrastructure layer 1110. In at least one embodiment, the resource manager 1126 may coordinate with the resource coordinator 1112 to manage these mapped or allocated computing resources.

[0114] In at least one embodiment, the software 1132 included in software layer 1130 may include software used by at least a portion of nodes CR1116(1)-1116(N), grouped computing resources 1114, and / or the distributed file system 1128 of framework layer 1120. One or more types of software may include, but are not limited to, internet web search software, email virus scanning software, database software, and streaming video content software.

[0115] In at least one embodiment, one or more applications 1142 included in application layer 1140 may include one or more types of applications used by at least a portion of nodes CR1116(1) to 1116(N), grouped computing resources 1114, and / or the distributed file system 1128 of framework layer 1120. One or more types of applications may include, but are not limited to, any number of genomics applications, cognitive computing applications, and machine learning applications, including training or inference software, machine learning framework software (such as PyTorch, TensorFlow, Caffe, etc.), or other machine learning applications used in conjunction with one or more embodiments.

[0116] In at least one embodiment, any of the configuration manager 1124, resource manager 1126, and resource coordinator 1112 can perform any number and type of self-modification actions based on any amount and type of data acquired in any technically feasible manner. In at least one embodiment, the self-modification actions can protect the data center operator of data center 1100 from making potentially undesirable configuration decisions and may prevent underutilization and / or poor performance of parts of the data center.

[0117] In at least one embodiment, data center 1100 may include tools, services, software, or other resources to train one or more machine learning models or to use one or more machine learning models to predict or infer information according to one or more embodiments described herein. In at least one embodiment, machine learning models can be trained by computing weight parameters according to a neural network architecture using the software and computing resources described above with respect to data center 1100. In at least one embodiment, trained machine learning models corresponding to one or more neural networks can be used to infer or predict information using the resources described above with respect to data center 1100 by using weight parameters computed through one or more training techniques. Deep learning can be advanced using any suitable learning network and the computing power of data center 1100. Thus, deep neural networks (DNNs), recurrent neural networks (RNNs), or convolutional neural networks (CNNs) can be supported simultaneously or concurrently using hardware in the data center. For example, once a network is trained and successfully evaluated to identify data within a subset or slice, the trained network can provide similar representative data for use with the collected data.

[0118] In at least one embodiment, the data center 1100 may use a CPU, application-specific integrated circuit (ASIC), GPU, FPGA, or other hardware to perform training and / or inference using the resources described above. For example, embodiments of this disclosure may be used to cool such hardware. Moreover, one or more of the software and / or hardware resources described above may be configured to allow users to train or perform services or any other artificial intelligence services that infer information such as pressure, flow rate, temperature, and location information.

[0119] Inference and / or training logic 1115 can be used to perform inference and / or training operations associated with one or more embodiments. In at least one embodiment, inference and / or training logic 1115 can be used in data center 1100 (whether for grouped computing resources 1114, for one or more nodes CR1116(1)-1116(N), or elsewhere) or in other systems described herein for inference or prediction operations based at least in part on weight parameters computed using neural network training operations, neural network functions and / or architecture, or neural network use cases described herein. In at least one embodiment, inference and / or training logic 1115 can include, but is not limited to, hardware logic where computing resources are dedicated or otherwise exclusively used in conjunction with weight values ​​or other information corresponding to one or more neuron layers within a neural network. In at least one embodiment, inference and / or training logic 1115 can be used with application-specific integrated circuits (ASICs) such as Tensorflow® processing units from Google, Graphcore, etc. TM This can be used in conjunction with an inference processing unit (IPU) or a Nervana® processor from Intel (such as "Lake Crest"). For example, embodiments of this disclosure can be used to cool such hardware.

[0120] In at least one embodiment, the inference and / or training logic 1115 may be used in conjunction with central processing unit (CPU) hardware, graphics processing unit (GPU) hardware, or other hardware such as a field-programmable gate array (FPGA), each of which may be cooled, for example, using embodiments of this disclosure. In at least one embodiment, the inference and / or training logic 1115 includes, but is not limited to, code and / or data storage modules that may be used to store code (such as graphical code), weight values, and / or other information, including bias values, gradient information, momentum values, and / or other parameter or hyperparameter information. In at least one embodiment, each of the code and / or data storage modules is associated with a dedicated computing resource. In at least one embodiment, the dedicated computing resource includes computing hardware that further includes one or more ALUs that perform mathematical functions, such as linear algebraic functions, solely on the information stored in the code and / or data storage modules, and the results are stored in the active storage module of the inference and / or training logic 1115.

[0121] The switches within each layer (e.g., edge layer, convergence layer, core layer) can be 1U switches. Switches can be electrical switches, optical switches, hybrid electro-optical switches, or any combination thereof. Switches can be implemented using suitable hardware and / or software that enables signal routing within appropriate domains. For example, an electrical switch may include a receiver that receives optical signals and converts them into electrical signals for routing within the electrical switch. The receiver of an electrical switch may include a transimpedance amplifier (TIA), a photodetector, and a controller, all used to convert optical signals into electrical signals. Each electrical switch may also include a transmitter that converts electrical signals routed within the electrical switch into optical signals for output to another (optical or electrical) switch within the system. For example, the transmitter of an electrical switch may include a light source, a modulator, and a controller that controls the modulator and the light source. In some embodiments, the receiver / transmitter pair may be integrated into a single transceiver. Each electrical switch may also include an internal switching circuitry system for routing electrical signals within the electrical switch.

[0122] A switch (whether an electrical switch, optoelectronic switch, and / or quantum switch) may include one or more input circuits and one or more output circuits connected via a switching core. In some embodiments, a switch may include multiple inputs and outputs.

[0123] Several such architectures have been proposed, including Next-Generation I / O (NGIO) and Future I / O (FIO), culminating in the InfiniBand architecture, which is being promoted by a consortium led by industry leaders including Intel, Sun, Hewlett Packard, IBM, Compaq, Dell, and Microsoft. Storage Area Networks (SANs) provide a similar packetized serial approach for high-speed storage access, which can also be implemented using the InfiniBand architecture.

[0124] Communication between parallel buses and packet networks typically requires a communication interface to convert bus cycles into appropriate packets and vice versa. For example, a host channel adapter or destination channel adapter can be used to link a parallel bus (such as a PCI bus) to an InfiniBand construct. When the adapter receives data from a device on the PCI bus, it inserts the data into the payload of the InfiniBand packet and then adds appropriate headers and error checking codes, such as Cyclic Redundancy Check (CRC) codes, as needed for network transmission. The InfiniBand packet header includes a routing header and a transport header. The routing header contains information at the data link protocol layer, including fields required for routing packets within and between subnets. The transport header contains higher-level end-to-end transport protocol information. Similar headers are used in other types of packet networks known in the art, such as Internet Protocol (IP) networks.

[0125] Figure 12The illustration shows an example network configuration whose components can be used to implement various aspects of different embodiments, such as providing, generating, modifying, encoding, processing, fusing, and / or transmitting generated image data, calculated measurements, or other such content. In at least one embodiment, client device 1202 can use components of content application 1204 on client device 1202, as well as data locally stored on the client device, to generate or receive data for a session. In at least one embodiment, content application 1224 executing on computer or processor 1220 (e.g., cloud server or control system) can initiate a session associated with at least one client device 1202 (e.g., vehicle or robot), such as using a session manager and user data stored in user database 1236, and can enable the selection and / or retrieval of content such as liquid coolant or server thermal data from storage 1234 for use by test module 1232 to calculate one or more performance metrics for monitoring module 1228, which can provide flow or thermal data to control the flow or temperature in an environment where the data is to be used to determine the appropriate operating conditions.

[0126] Content manager 1226 can work with at least these different modules to perform tests and analyses, and may instruct any actions to be taken in response to performance metrics failing to meet operational requirements. At least a portion of the data or instruction content can be transmitted to client device 1202 and / or physical device 1270 using a suitable transmission manager 1222 for sending via download, streaming, or other such transmission channels. An encoder can be used to encode and / or compress at least a portion of the data before it is transmitted to client device 1202. In at least one embodiment, client device 1202 receiving such content can provide this content to a corresponding content application 1204, which may also or optionally include a graphical user interface 1210, a traffic monitoring module 1212, and a control module 1214 for providing, compositing, rendering, combining, modifying, or using the content on or through client device 1202 for presentation, navigation, control (or other purposes), such as transmission to physical device 1270.

[0127] In some embodiments, the computer / processor 1220 and the client device 1202 may be able to communicate directly without transmitting data over the network 1240 to avoid issues such as latency and availability. A decoder may also be used to decode data received over the network 1240 for presentation by the client device 1202, such as presenting image content or performance metrics via a display device 1206, and playing audio (e.g., corresponding sounds or synthesized speech) via at least one audio playback device 1208 (e.g., a speaker or headphones).

[0128] In at least one embodiment, at least a portion of the content may already be stored on, rendered on, or accessible to client device 1202, so that at least a portion of the content does not need to be transmitted over network 1240. For example, the content (e.g., hot data) may have been pre-downloaded or stored locally on a hard disk or optical disk. In at least one embodiment, the content may be transmitted from computer / processor 1220 or user database 1236 to client device 1202 using a transmission mechanism such as data streaming. In at least one embodiment, at least a portion of the content may be obtained, enhanced, and / or streamed from other sources (e.g., third-party service 1260 or other client devices 1250), which may also include content applications for generating, updating, enhancing, or providing map content. In at least one embodiment, a portion of this functionality may be executed using multiple computing devices, or multiple processors in one or more computing devices, for example, a combination of CPU and GPU (graphics processing unit).

[0129] In at least some of these examples, the client device may include any suitable computing device, such as a desktop computer, laptop computer, set-top box, streaming media device, game console, smartphone, tablet computer, VR headset, AR glasses, wearable computer, or smart TV. Each client device may submit requests via at least one wired or wireless network, such as the Internet, Ethernet, local area network (LAN), or cellular network.

[0130] In this example, these requests may be submitted to an address associated with a cloud provider that operates or controls one or more electronic resources within the cloud provider's environment, such as a data center or server farm. In at least one embodiment, the requests may be received or processed by at least one edge server located at the network edge and outside at least one security layer associated with the cloud provider's environment. This reduces latency and improves the security of resources within the cloud provider's environment by allowing client devices to interact with a closer server.

[0131] In at least one embodiment, the system can be used to monitor or manage the thermal status of a server that includes a cold plate as a liquid manifold. In other embodiments, the system can be used for other purposes, such as providing control over the flow of liquid coolant or performing deep learning operations. In at least one embodiment, the system can be implemented using an edge device or may include one or more virtual machines (VMs). In at least one embodiment, the system can be implemented at least partially in a data center or at least partially using cloud computing resources.

[0132] According to at least some embodiments of this disclosure, the following can be used: Figure 13 The exemplary data center 1300 shown here has the improved cooling system described herein. Any or more of the features described above can be used in a data center such as data center 1300. For a deeper understanding, many specific details are listed herein, but the concepts presented herein can be practiced even without considering one or more of these specific details.

[0133] In at least one embodiment, the data center cooling system can respond to sudden surges in heat demand caused by the constantly changing computing load in today's computing components. In at least one embodiment, because these demands vary, or the range of cooling demands may vary from minimum to maximum, a suitable cooling system must be used to meet these demands economically. In at least one embodiment, a liquid cooling system can be used for medium to high cooling demands. In at least one embodiment, partial immersion cooling can economically meet higher cooling demands. In at least one embodiment, these different cooling demands also reflect different thermal characteristics of the data center. In at least one embodiment, the heat generated by these components, servers, and racks is collectively referred to as thermal characteristics or cooling demands, because cooling demands must fully address thermal characteristics.

[0134] In at least one embodiment, a split-type heatsink system is disclosed. In at least one embodiment, this split-type heatsink system addresses the heat feature in associative computing or data center devices, including graphics processing units (GPUs), central processing units (CPUs), switches, and dual in-line memory modules (DIMMs). In at least one embodiment, these components may be referred to herein as high-heat-density computing components or simply high-power components. Furthermore, in at least one embodiment, the associative computing or data center device may be a processing card having one or more GPUs, switches, or CPUs. In at least one embodiment, each of the GPU, switch, and CPU may be a heat-generating feature of the computing device. In at least one embodiment, the GPU, CPU, or switch may have one or more cores, and each core may be a heat-generating feature.

[0135] In at least one embodiment, a split heatsink as described above (and more specifically, a central heatsink or a first heatsink section as described above) can be thermally coupled to one or more high-power components as described above. In some embodiments, a liquid cooling system can be used as an alternative to or complement to the split heatsink (or only in conjunction with other data center components besides the split heatsink).

[0136] In at least one embodiment, the following can be used: Figure 13 The exemplary data center 1300 shown herein has a cooling system that can be improved as described herein. In at least one embodiment, the data center 1300 may be or include one or more server rooms 1302, each housing racks 1310 and auxiliary equipment for accommodating one or more servers on one or more server bays. In at least one embodiment, the data center 1300 is supported by a cooling tower 1304 located outside the data center 1300. In at least one embodiment, the cooling tower 1304 dissipates heat from inside the data center 1300 by acting on a main cooling loop 1306. In at least one embodiment, a cooling distribution unit (CDU) 1312 is provided between the main cooling loop 1306 and a second or auxiliary cooling loop 1308 to absorb heat from the second or auxiliary cooling loop 1308 back to the main cooling loop 1306. In at least one embodiment, in one aspect, the auxiliary cooling loop 1308 may have various pipes connected to the server bays as needed. In at least one embodiment, loops 1306 and 1308 are shown in line diagram form, but those skilled in the art will understand that one or more pipe structures may be used. In at least one embodiment, flexible polyvinyl chloride (PVC) pipes and associated conduits may be used to move fluid in each provided loop 1306, 1308. In at least one embodiment, one or more coolant pumps may be used to maintain pressure differentials within coolant loops 1306, 1308 to move coolant based on temperature sensors located at different locations, including within a server room, within one or more racks 1310, and / or within server chassis or server racks within one or more racks 1310.

[0137] In at least one embodiment, the coolant in the main cooling circuit 1306 and the auxiliary cooling circuit 1308 may be at least water and an additive. In at least one embodiment, the additive may be ethylene glycol or propylene glycol. In operation, in at least one embodiment, the main cooling circuit and the auxiliary cooling circuit may each have their own coolant. In at least one embodiment, the coolant in the auxiliary cooling circuit may be dedicated to the needs of components in the server bay or associated rack 1310. In at least one embodiment, the CDU 1312 is capable of fine control over the coolant provided in the coolant circuits 1306, 1308, either independently or simultaneously. In at least one embodiment, the CDU may be adapted to control the flow rate of the coolant so that the coolant is properly distributed to absorb heat generated within the associated rack 1310. In at least one embodiment, a more flexible conduit 1314 is provided from the auxiliary cooling circuit 1308 to each server bay to provide coolant to the electrical and / or computing components therein.

[0138] Heat transfer fluids typically include water, aqueous solutions (e.g., propylene glycol-water), brine, antifreeze, mixtures of antifreeze and water, oil, alcohol, mercury or the like, or any other suitable heat-conducting fluid. The heat transfer fluid can be a conductive coolant and can include water, deionized water, or a refrigerant (e.g., R-134a), a mixture of water and additives (e.g., a mixture of water and ethylene glycol or a mixture of water and propylene glycol, such as 25% propylene glycol in deionized water). The heat transfer fluid can also be a single dielectric fluid (e.g., water-free for the purposes of this disclosure), or water combined with an additive containing at least one dielectric fluid, such as one or more of deionized water, ethylene glycol, and propylene glycol. In at least one embodiment, the heat transfer fluid can be an absorption chiller whose working fluid is a mixed solution containing lithium bromide as the absorbent material and water as the carrier material. The heat transfer fluid can also be a two-phase coolant with a boiling point below the expected operating temperature of the electronic device. Exemplary two-phase coolants include 2,3,3,3-tetrafluoropropylene, 1,1,1,2-tetrafluoroethane, and water.

[0139] In at least one embodiment, the conduit 1318 forming part of the auxiliary cooling circuit 1308 may be referred to as a server room manifold. Furthermore, in at least one embodiment, an additional conduit 1316 may extend from the manifold conduit 1318 and may also form part of the auxiliary cooling circuit 1308, but may be referred to as a manifold. In at least one embodiment, the coolant conduit 1314 enters the rack as part of the auxiliary cooling circuit 1308, but may be referred to as a rack cooling manifold within one or more racks. In at least one embodiment, the manifold 1316 extends to all racks along a row in the data center 1300. In at least one embodiment, the piping of the auxiliary cooling circuit 1308, including coolant manifolds 1318, 1316, and 1314, may be improved by at least one embodiment herein. In at least one embodiment, a cooler 1320 may be provided in the main cooling circuit within the data center 1300 to support cooling before the cooling tower. In at least one embodiment, an additional cooling circuit, which may exist in the main control circuit and provide external cooling to both the rack exterior and the auxiliary cooling circuit, may be integrated with the main cooling circuit and distinguished from the auxiliary cooling circuit in this disclosure.

[0140] In at least one embodiment, during operation, heat generated within the server trays of the provided rack 1310 can be transferred via the flexible conduit of the manifold 1314 of the second cooling circuit 1308 to coolant leaving one or more racks 1310. In at least one embodiment, a second coolant from CDU 1312 (in the auxiliary cooling circuit 1308) is used to cool the provided rack 1310 and flows through the provided conduit to one or more racks 1310. In at least one embodiment, the second coolant from CDU 1312 flows out from one side of a data center manifold having conduit 1318, flows through manifold 1316 to one side of the rack 1310, and flows through another conduit 1314 through one side of the server tray. In at least one embodiment, used or returned second coolant (or departing second coolant carrying heat from computing components) flows out from the other side of the server tray (e.g., entering from the left side of the rack and exiting from the right side of the server tray after circulating through the server tray or components on the server tray). In at least one embodiment, used second coolant exiting the server rack or cabinet 1310 flows from a different side (e.g., the exit side) of pipe 1314 and enters a parallel but also exit side of manifold 1316. In at least one embodiment, used second coolant enters from manifold 1316 into a parallel portion of data center manifold 1318, flowing in the opposite direction to the incoming second coolant (which may also be a replaced second coolant), and flows toward CDU 1312.

[0141] In at least one embodiment, the used second coolant exchanges heat with the main coolant in the main cooling circuit 1306 via CDU 1312. In at least one embodiment, the used second coolant can be refreshed (e.g., at a relatively lower temperature compared to the temperature of the used second coolant stage) and prepared to be circulated back to one or more computing components via the second cooling circuit 1308. In at least one embodiment, various flow and temperature control features in CDU 1312 are capable of controlling the heat exchanged from the used second coolant, or controlling the flow rate of the second coolant into and out of CDU 1312. In at least one embodiment, CDU 1312 is also capable of controlling the flow rate of the main coolant in the main cooling circuit 1306.

[0142] Even in the combination of the above Figure 13 In the liquid-cooled data center 1300 described, split-type heat sinks, such as split-type heat sink 100, can also be used to facilitate cooling of high-power components and adjacent or nearby low-power components, which can particularly advantageously improve the maintainability of adjacent or nearby low-power components. Removing a single section of the split-type heat sink 100 to maintain underlying components is more efficient than removing, for example, the entire liquid-cooled heat sink or cold plate (such as combined...). Figure 13 The described heat sink or cold plate is much easier to use.

[0143] Now go to Figure 14In at least one embodiment, the neural network can be embedded in various devices, enabling these devices to predict future states and proactively adjust themselves, at least in part, based on these predictions. In at least one embodiment, this can include directly mounting neural network-enabled devices within a liquid-cooled server rack 1400. In at least one embodiment, rack 1400 can include multiple liquid-cooled servers 1402 or other such devices. In at least one embodiment, rack 1400 may include a rack manifold 1404 supporting the neural network, which provides a liquid flow to each liquid-cooled server 1402 via an inlet valve 1408 and returns the liquid with heat removed from that server 1402 via an outlet valve 1410. In at least one embodiment, rack manifold 1404 may include multiple auxiliary liquid distribution control panels 1406 supporting the neural network. In at least one embodiment, one such control panel 1406 may be provided for each inlet valve 1408 and outlet valve 1410 for each server. In at least one embodiment, rack 1400 may further include a rack power distribution unit (PDU) 1412 supporting neural networks. The PDU 1412 may contain multiple ports designed to distribute power to computers or networking devices within rack 1400. In at least one embodiment, each port of PDU 1412 may be associated with a power distribution control board 1414 supporting neural networks. In at least one embodiment, sensors may capture information about the computing environment inside and / or outside rack 1400 (including inside and / or outside any individual server 1402 located therein), including temperature, airflow, or other such aspects. In at least one embodiment, liquid cooling may remove a certain amount of heat from server 1402, but temperatures at different locations may vary due to factors such as load variations and external temperature fluctuations, and may reach or exceed temperature limits that allow these devices to continue operating normally. In at least one embodiment, efforts may be made to ensure that temperatures at specific locations remain below acceptable limits, where these locations may be related to the junction or core temperatures of processors (e.g., CPUs or GPUs), memory modules, or power supplies.

[0144] In at least one embodiment, the neural networks in these control boards 1406, 1414 can be responsible for predicting values ​​such as coolant flow rate or airflow rate, and the temperature of these various locations at one or more future points in time. In at least one embodiment, the corresponding boards, manifolds, PDUs, or other such devices can be adjusted using the predictions from these networks to ensure appropriate temperature and flow rates and to ensure proper device operation. In at least one embodiment, these adjustments can include adjusting the amount of fluid or air flowing into or out of servers, radiators, or other devices, and the temperature of that fluid or air. In at least one embodiment, this can also include adjusting the amount of power supplied to one or more of these devices and / or the fans controlling airflow around these devices. In at least one embodiment, these adjustments can include adjustments similar to those made in other data centers, but these adjustments can be based on future predictions rather than observed states, thus enabling the system to operate proactively rather than reactively. In at least one embodiment, such monitoring can be continuous, or at least periodically, to ensure continuous and normal operation.

[0145] In at least one embodiment, each neural network in each component can be specifically trained for that component type and can receive updated network parameters generated as a result of further training or continuous learning. In at least one embodiment, inference results generated by the individual devices can also be shared with other devices in the data center, which helps to make more accurate predictions. In at least one embodiment, each control board 1406, 1414 can be a neural network-enabled board, ASIC, or other such component. In at least one embodiment, the control board can include components of a general-purpose computer, including at least one processor (e.g., Figure 10 CPU 1002 and memory (e.g.) Figure 10The main memory 1004 and related circuitry are used. In at least one embodiment, NVIDIA's Jetson boards, which are complete system modules (SPMs) containing a CPU, GPU, PMIC, DRAM, and flash memory, can be used. In at least one embodiment, such modules can also be expanded to provide additional functionality or capabilities. In at least one embodiment, each such board can operate as a small, AI-enabled or machine learning-enabled computer. In at least one embodiment, multiple such boards can be used in parallel, for example on rack manifold 1404, to concurrently process data from multiple high-resolution sensors. In at least one embodiment, these different neural networks can make predictions that can be shared between servers, racks, or data centers, for example, to provide accurate predictions for various locations at one or more future points in time. In at least one embodiment, appropriate adjustments can then be made as needed to maintain temperature and other parameters at appropriate levels or values. In at least one embodiment, the liquid distribution control board 1406 can make predictions that can be used to adjust control of relevant flow values ​​only, or these predictions can be shared to allow for additional adjustments based at least in part on these predictions. In at least one embodiment, the control system for the data center can collect predictions from these different boards or networks to make adjustments more suitable for the entire data center. In at least one embodiment, adjustments can be made at the server, rack, pod, data center, or other similar level.

[0146] In at least one embodiment, a data center-class rack liquid cooling method enabled by neural networks can utilize... Figure 15A The components of the system 1500 shown. In at least one embodiment, the external cooling unit 1502 can provide liquid at a determined temperature to the data center cooling distribution unit 1504. In at least one embodiment, the unit 1504 may include a set of neural network-enabled distribution unit control boards 1506 that can perform inference and adjust based on inferred future values. In at least one embodiment, this may include adjusting the temperature of the liquid to be received from the external cooling unit 1502, or adjusting the flow rate of the liquid into and / or out of the external unit 1502. In at least one embodiment, this may also include adjusting the flow rate into manifolds 1508 for a row of liquid-cooled racks 1512, and the flow rates into and out of these racks, as can be determined at least in part based on the inference results of the neural network-enabled control boards 1510 built into the manifolds 1510. In at least one embodiment, the flow rate levels into and out of different liquid-cooled racks may be different. In at least one embodiment, the flow rate into each server in the rack may also be different, as combined with Figure 14As described. In at least one embodiment, a similar method can also be used in air-cooled data centers. In at least one embodiment, temperature and airflow can be measured, and future values ​​can be predicted, for example, over the next 30 seconds, 1 minute, 5 minutes, or 1 hour, thereby adjusting air temperature, airflow, power distribution, and other relevant aspects. In at least one embodiment, these predictions can be made using various types of input data, such as server load, current temperature, current pressure, flow rate, power consumption, and other relevant data from different locations within the data center.

[0147] In at least one embodiment, these neural network-enabled boards can control air or liquid flow, supply, and return flow at different locations (e.g., racks, servers, or GPU levels). In at least one embodiment, these boards can be used to predict, control, and create optimal solutions for data centers to maintain required operating conditions and avoid equipment loss or downtime due to temperature issues or thermal events. In at least one embodiment, this can be based on current and anticipated loads and measured or detected network conditions. In at least one embodiment, this approach can also be applied to other types of devices, such as UPS power supplies. In at least one embodiment, any device with logic boards or components can be injected with such small neural network code to achieve the above and other similar purposes. In at least one embodiment, such a network can access all available devices in a unit or device, such as in devices like PDUs, where boards can adjust or completely cut off power to server nodes, for example, in the event of a detected or predicted leak where continued operation could damage the associated equipment. In at least one embodiment, operation can be gracefully terminated or power turned off to avoid damage. In at least one embodiment, an executing program can be terminated to at least prevent data loss.

[0148] In at least one embodiment, the cooling system may employ a direct recirculation design, such as... Figure 15AAs shown. In at least one embodiment, coolant from CDU 1504 is directed along supply manifold 1508 to the individual server racks. In at least one embodiment, the first server rack (A) will receive liquid first along this flow, while the last server rack (E) will receive liquid last along this flow. In at least one embodiment, return manifold 1510 will be used to receive the heated liquid passing through these racks and return the heated liquid to CDU 1504. In at least one embodiment, the heated liquid returning from the first server rack (A) has the shortest return path, while the heated liquid returning from the last server rack (E) has the longest return path. In at least one embodiment, this results in the overall liquid path 1514 of the first server rack (A) being much shorter than the overall liquid path 1516 of the last server rack (E) 1516. In at least one embodiment, factors such as pressure and flow rate may vary significantly throughout the cooling system loop. In at least one embodiment, this may include lower pressure near the ends of the cooling loop and higher pressure in the middle of the loop. In at least one embodiment, this unbalanced design can lead to unacceptable variations in flow rates in data centers or other environments because different path lengths result in varying pressures, temperatures, and other environmental factors that affect these flow rates. In at least one embodiment, the different amounts of time required for the liquid to flow along these different paths also contribute to these flow rate variations.

[0149] In at least one embodiment, the following can be employed: Figure 15B The system 1520 is based on a reverse flow design. In at least one embodiment, the system includes a reverse flow manifold 1522 that first receives heated liquid from a first server rack (A) and finally receives heated liquid from a last server rack (E), but instead of directly guiding the flow back to CDU 1506, it allows the liquid to flow from the first server rack to the last server rack (e). Figure 15B As shown (from right to left), the fluid for each server rack will flow at least through the corresponding manifold along a path 1524 of similar length. In at least one embodiment, these path lengths may be virtually equal, allowing for slight variations depending on factors such as layout and component type. In at least one embodiment, the fluid for each server rack will flow through almost the entire length of the supply manifold and the return manifold, rather than as shown... Figure 15AUnlike a direct return design, the liquid flows through different sections. In at least one embodiment, this ensures that the liquid in all server racks experiences the same variations in temperature, pressure, and other environmental parameters throughout the manifold's operation. In at least one embodiment, this balanced design helps reduce flow rate variations of the liquid delivered to different locations throughout the cooling system. In at least one embodiment, this helps ensure that flow rate variations remain within the desired range or below a maximum variation threshold. In at least one embodiment, this balanced design helps ensure that the individual cooling components of the liquid cooling system deliver equal and sufficient amounts of liquid to all servers and racks in the data center.

[0150] In at least one embodiment, all supply and return fluid lines between one or more CDUs and one or more direct-to-chip (D2C) cold plate cooling loops or heat exchangers in immersion liquid-cooled blade servers can employ reverse recirculation, rather than direct recirculation, in various data center manifolds, exhaust manifolds, and rack manifolds. In at least one embodiment, electronic valve controls at each outlet of these manifolds, exhaust manifolds, and rack manifolds are capable of monitoring and controlling the flow rate of primary and / or auxiliary fluids into and out of these CDUs, which lead to cold plates attached to components such as CPUs, GPUs, and switches, as well as other D2C liquid-cooled components and immersion blade servers. In at least one embodiment, the liquid cooling system can employ a self-balancing flow design in all primary and auxiliary loops, combined with electronic monitoring and control utilizing AI or machine learning, based at least in part on data received from various flow rate, pressure, temperature, power, and other environmental or operational sensors. In at least one embodiment, this approach can improve or optimize liquid distribution to liquid-cooled components such as servers and processors in environments such as data centers.

[0151] In at least one embodiment, neural network-based flow control can be used with this design to provide further variation minimization or flow consistency. In at least one embodiment, the combination of self-balancing with electronic monitoring and control can provide a highly balanced and optimized coolant flow for each of servers, racks, processors, or other such components, without these components competing with each other or requiring complex control schemes. In at least one embodiment, flow controllers can be located at various points throughout the cooling system, for example, to guide coolant inflow and outflow from components such as racks, servers, and electronic components, where the liquid can be supplied via paths such as manifolds and rack manifolds. In at least one embodiment, these flow controllers can implement incremental control (e.g., partial opening and closing) to provide an accurate self-balancing approach. In at least one embodiment, a self-balancing design with finely adjustable control can provide accurate balancing of liquid flow at each (if not all) outlet of the liquid cooling system. In at least one embodiment, this method can be used to provide, for example, a flow rate of 50 liters per minute (lpm), with a tolerance, threshold, or range of variation of + / - 5 lpm. In contrast, a design without this method may have a flow rate of approximately 60 lpm at one location and approximately 40 lpm at another location due to differences such as pressure gradients and gravity effects, which could be an unacceptable flow imbalance in the entire system.

[0152] In at least one embodiment, the control system 1540 can be as follows: Figure 15CThe implementation is shown. In at least one embodiment, a reverse-flow manifold design can be used to provide pressure and flow balance throughout the system. In at least one embodiment, a controllable flow valve 1546 can be used at various inlet and / or outlet locations, such as inlets and outlets from the manifold to individual racks. In at least one embodiment, the flow manager 1542 can receive data from various sensors, devices, or data sources in the relevant environment. In at least one embodiment, the flow manager 1542 can provide at least a portion of this data as input to one or more neural networks (e.g., deep neural networks (DNNs)) of the inference module 1544. In at least one embodiment, the sensors and valves can be calibrated for the type of heat transfer fluid used. In at least one embodiment, there can be one flow manager 1542 for the entire data center or environment, or multiple flow managers can be set for individual rows, racks, or servers. In at least one embodiment, neural networks can also be used at each individual flow control valve, such as those communicating with one or more independent flow controllers. In at least one embodiment, the flow manager 1542 can receive one or more inference results from the inference module 1544, wherein these inference results can include adjustments or settings for the individual flow control valves. In at least one embodiment, the flow manager 1542 may send this data or related data or instructions to these individual flow control valves 1546 or other such components for minor adjustments to help balance the flow across the entire system.

[0153] In at least one embodiment, this method can be used at the rack level or server level, such as... Figure 15D The system 1560 is shown in the diagram. In at least one embodiment, the liquid-cooled rack may have a rack manifold with a reverse reflux design, thereby ensuring that the liquid path lengths for each server in the rack are equal. In at least one embodiment, the size of the manifold may be much smaller than the size of the corresponding rack, such as... Figure 15D The exploded view shown is provided for clarity. In at least one embodiment, the flow manager 1562 may utilize the inference module 1564 to determine the necessary adjustments to the various flow control valves 1566 within the rack to balance flow between the servers in the rack. In at least one embodiment, the flow manager may also provide such instructions to flow control valves or components within the servers, for example, to direct coolant to specific cold plates, processors, or other internal server components. In at least one embodiment, using these intelligent control valves can provide well-balanced piping distribution and controllability. In at least one embodiment, each of these intelligent control valves 1566 can be fine-tuned as needed, such as under the guidance of the flow manager. In at least one embodiment, these flow valves may employ PID or PIV servo control, where PIV allows control based on both position and velocity errors.

[0154] In at least one embodiment, one or more neural networks can be trained to infer the amount of liquid cooling required for various components under various conditions. In at least one embodiment, this may include determining how much additional cooling might be needed when the GPU's operating power increases from a more typical 300 watts to 500 watts, which might require 1.7 lpm instead of 1.0 lpm. In at least one embodiment, the neural network can infer adjustments to the liquid flow rate of the component and can enable proactive and accurate adjustments to avoid significant temperature fluctuations. In at least one embodiment, one or more operation tables may exist that correlate factors such as proportional flow rate with power consumption; these operation tables can be used to train these models or to determine adjustments based on inferred state data. In at least one embodiment, these tables can also be used for different types of components or different models of the same type of component to account for the inherent differences between these components. In at least one embodiment, inference can be performed at different levels of granularity, from the entire data center level to the level of internal server components (e.g., GPUs). In at least one embodiment, data may be collected by a component such as a board-level management controller (BMC) or a server and provided to a flow manager via a network port; the flow manager can send instructions to one or more flow controllers to make one or more appropriate adjustments.

[0155] In at least one embodiment, the neural network can predict the fluid state at various locations at one or more future times, at least in part, based on available information. In at least one embodiment, this information may include snapshots of current information or may include at least some recent data. In at least one embodiment, these predictions can be used to adjust aspects such as flow rate. In at least one embodiment, the cooling distribution unit can adjust the liquid flow rate or coolant temperature. In at least one embodiment, such devices can utilize these predictions and proactively take measures to prevent any adverse events before they occur. In at least one embodiment, various sensors can provide information about the current state of the computing environment. In at least one embodiment, this may include the use of sensors such as temperature sensors, load sensors, flow sensors, or pressure sensors, whose data may be acquired instantaneously or historically. In at least one embodiment, predictions based on data from these sensors can be compared with one or more thresholds, ranges, or other operating criteria to determine whether any changes are needed. In at least one embodiment, this may include adjustments to prevent unacceptable temperature rises at specific locations in a data center or other similar environment. In at least one embodiment, this may include remedial measures such as closing valves or increasing coolant flow, adjusting coolant temperature, or issuing alarms.

[0156] In integrated circuit (IC) chip packaging, various packaging schemes are employed, including traditional two-dimensional (2D) IC packaging and the recently introduced 2.5D and 3D IC packaging. In 2D IC packaging, multiple chips are mounted on a printed circuit board, with high-performance logic, lower-performance logic, memory, analog / radio frequency (RF) functions, and other functional components placed as discrete devices in separate chip packages. In contrast, in 2.5D and 3D IC packaging, multiple IC chips are mounted on a silicon interposer, rather than a conventional packaging substrate. The silicon interposer is typically a silicon wafer, and because the manufacturing process used to form the conductive traces is the same as the process used to form the metal interconnects in the metallization layer of the silicon chip, this silicon interposer allows for the formation of very small and high-density conductive traces between multiple IC chips.

[0157] Compared to 2.5D IC packages and 3D IC packages, circuit boards using separately packaged chips (such as 2D IC packages) have several disadvantages. For example, 2D IC packages are typically larger, heavier, and consume more power, and their speed is lower than equivalent 2.5D or 3D IC packages because signals propagate relatively slowly from one chip to another on the circuit board. Furthermore, 2D IC packages have more potential points of failure because solder joints on the circuit board are more likely to fail than electrical connections formed in the interposer. That said, troubleshooting 2D IC packages is relatively simple once different chips are mounted on the circuit board. In particular, the conductive traces carrying I / O signals between the individual chips on the circuit board are readily accessible and can be used to measure specific I / O signals during troubleshooting.

[0158] In contrast, troubleshooting 2.5D or 3D IC packages is far more difficult because I / O signals transmitted between different chips are typically embedded in the silicon interposer, making them physically inaccessible. Furthermore, due to the high bandwidth and dense density of 2.5D and 3D IC packages, typical implementations may involve routing thousands of conductive traces between different chips. An example of such an implementation is a memory bus between a processor and a high-bandwidth memory chip. In this implementation, even if these traces could be physically accessed through the silicon interposer using probes, accurately and reliably selecting specific conductive traces or combinations of conductive traces for troubleshooting the IC package would be extremely difficult, if not impossible.

[0159] In at least one embodiment, one or more parallel processors (such as parallel processing unit 1014 of parallel processing system 1012) include circuitry optimized for graphics and video processing (including, for example, video output circuitry) and constitute a graphics processing unit (“GPU”). In at least one embodiment, one or more parallel processors include circuitry optimized for general-purpose processing. In at least one embodiment, components of a computing system (such as computer system 1000) may be integrated with one or more other system elements on a single integrated circuit. For example, in at least one embodiment, one or more parallel processors, a memory hub (such as main memory 1004 of computer system 1000), one or more processors (such as CPU 1002 of computer system 1000), and an I / O hub (which facilitates communication with one or more display devices 1006, one or more input devices 1008, and / or network interface 1022 of a system such as computer system 1000) may be integrated into a system-on-a-chip (SoC) integrated circuit. In at least one embodiment, components of such a computing system can be integrated into a single package to form a system-in-package (“SIP”) configuration. In at least one embodiment, at least some components of such a computing system can be integrated into a multi-chip module (“MCM”), which can be interconnected with other multi-chip modules to form a modular computing system. In at least one embodiment, an I / O subsystem and any display device are omitted from such a computing system.

[0160] Figure 16 This is a block diagram illustrating a computing system 1600 (e.g., a data center or high-performance computing (HPC) cluster) according to embodiments described herein. According to at least one embodiment, system 1600 includes multiple subsystems, such as multiple processing devices, multiple network devices, and multiple networks coupled to each other. The computing system 1600 is designed with multiple integrated circuits (referred to as processing devices), each of which may include one or more CPUs and GPUs, forming a powerful and flexible architecture.

[0161] Various processing devices are interconnected via NVLink or other high-speed interconnects to enable high-speed communication between subsystems, and are also connected via NICs or DPUs to ensure efficient data transmission on computing system 1600 and with one or more external networks 1630, 1636. In this example, system 1600 includes: a packet switch 1648 connecting NIC / DPU 1628 to network 1630; and a packet switch 1650 connecting NIC / DPU 1632 to network 1636.

[0162] Seamless data exchange and parallel processing are achieved through NVLink-coupled processing devices, thereby improving overall computing performance. The processing devices connect to multiple networks via one or more Network Interface Controllers (NICs) or Data Processing Units (DPUs), enabling the system to handle complex multi-network tasks with high bandwidth and low latency. This configuration is highly suitable for demanding applications requiring massive processing power, such as artificial intelligence (AI), machine learning (ML), and data-intensive computing, while ensuring robust connectivity and scalability across diverse network environments. The integrated circuits of the Computing System 1600 may include one or more CPUs and one or more GPUs.

[0163] Figure 16 An exemplary architecture of a multi-GPU architecture is also illustrated. As shown in the figure, computing system 1600 includes a processing device 1602 with a multi-GPU architecture. Specifically, processing device 1602 may be a system-on-a-chip and includes multiple subsystems, such as CPU 1606, GPU 1608, and GPU 1610. CPU 1606 may be coupled to GPU 1608 via die-to-die (D2D) or chip-to-chip (C2C) interconnect 1612 (e.g., ground reference signaling interconnect (GRS interconnect)). CPU 1606 may be coupled to GPU 1610 via D2D or C2C interconnect 1614. CPU 1606 may also be coupled to GPU 1608 and GPU 1610 via PCIe interconnect.

[0164] The CPU 1606 can be coupled to one or more NICs or DPUs, which in turn are coupled to one or more networks. For example, as Figure 16 As shown, CPU 1606 is coupled to a first NIC / DPU 1626, which is coupled to network 1630. CPU 1606 is also coupled to a second NIC / DPU 1628, which is coupled to network 1630 via switch 1648. NIC / DPU 1626 and NIC / DPU 1628 can be coupled to network 1630 via Ethernet (ETH), NVLINK, or InfiniBand (IB) connections.

[0165] The computing system 1600 also includes a processing device 1604 with a multi-GPU architecture. Specifically, the processing device 1604 includes multiple subsystems, including a CPU 1616, a GPU 1618, and a GPU 1620. The CPU 1616 can be coupled to the GPU 1618 via a D2D or C2C interconnect 1622. The CPU 1616 can be coupled to the GPU 1620 via a D2D or C2C interconnect 1624. The CPU 1616 can also be coupled to the GPU 1618 and GPU 1620 via a PCIe interconnect. The CPU 1616 can be coupled to one or more NICs or DPUs, which in turn are coupled to one or more networks. For example, as... Figure 16 As shown, CPU 1616 is coupled to a first NIC / DPU 1634, which is coupled to network 1636. CPU 1616 is also coupled to a second NIC / DPU 1632, which is coupled to network 1636 via switch 1650. NIC / DPU 1632 and NIC / DPU 1634 can be coupled to network 1636 via Ethernet (ETH), NVLINK, or InfiniBand (IB) connections.

[0166] In at least one embodiment, processing device 1602 and processing device 1604 can communicate with each other via NIC / DPU 1638 (e.g., via PCIe interconnect). Processing device 1602 and processing device 1604 can also communicate with each other via high-bandwidth communication interconnect 1640 (e.g., NVLink interconnect or other high-speed interconnect). Figure 16 The packet switches in the diagram can include, for example, Nvidia Quantum-2 switches. The NIC / DPU in the diagram can include, for example, Nvidia Bluefield DPUs.

[0167] Cooling systems (such as those mentioned above) Figure 1A-1B The system 100 includes a cooling system having one or more bases (such as base 128) with contact surfaces (such as contact surface 116) and at least one pressure surface (such as pressure surface 120) that can be used to cool one or more components of the system 1600. For example, such a cooling system can be used to cool one or more of CPUs 1606 and 1616, one or more of GPUs 1608, 1610, 1618 and 1620, an entire superchip including CPU 1606 and GPUs 1608 and 1610 and / or including CPU 1616 and GPUs 1618 and 1620, and any other physically heat-generating components in the system 1600.

[0168] Figure 17 An example computing environment 1700 according to at least one embodiment of the present disclosure is illustrated, wherein forward pass-off to available memory can be performed according to at least one embodiment. It should be understood that embodiments of the present disclosure can also be used with reference to alternative environments, and specific discussions of components are provided in a non-limiting manner and may include equivalents. Furthermore, for clarity and brevity, several features have been omitted. Additionally, the system and method can be used with a variety of different architectures. The example computing environment 1700 may include a server 1702, which can be used to perform HPC workloads, such as AI training or machine learning model training. In embodiments, server 1702 may be an application instance or a compute node. Server 1702 may include a CPU 1710 associated with a switch 1720 (e.g., a Peripheral Component Interconnect Fast (PCIe) switch), which can control at least some data transmissions on communication paths interconnecting various components. In embodiments, CPU 1710 may include a root complex processor.

[0169] PCIe switch 1720 may also be associated with GPU 1730 and DPU 1740, and may transfer data between at least some of CPU 1710, GPU 1730, DPU 1740, and other components. In one embodiment, PCIe switch 1720 may be associated with more than one GPU or more than one DPU. In another embodiment, PCIe switch 1720 may be located within DPU 1740. PCIe switch 1720 may manage the transfer of at least some of the data between CPU 1710, GPU 1730, and DPU 1740. In another embodiment, the number of GPUs associated with PCIe switch 1720 may be equal to the number of DPUs associated with PCIe switch 1740. In at least one embodiment, server 1702 may include, but is not limited to, any number of CPU 1710, PCIe switch 1720, GPU 1730, and / or DPU 1740, in any combination. For example, in at least one embodiment, server 1702 may include eight, sixteen, thirty-two, and / or more GPUs 1730. In at least one embodiment, for Figure 17 The communication paths that interconnect the various components (including but not limited to CPU 1710, PCIe switch 1720, GPU 1730 and DPU 1740) can be implemented using any suitable protocol, such as peripheral component interconnect (PCI) based protocols (e.g. PCIe), or other bus or point-to-point communication interfaces and / or protocols, such as NV-Link high-speed interconnect, or interconnect protocols.

[0170] DPU 1740 may include a network interface controller (NIC) 1742, DDR memory 1744, and a non-volatile memory fast (NVMe) device 1746. NIC 1742 is capable of interfaced with network 1704, which may also interface with additional NVMe devices available to DPU 1740, for example, via a structure. In one embodiment, DPU 1740 may not include NVMe device 1746. In another embodiment, NVMe device 1746 may reside on server 1702 instead of DPU 1740. In yet another embodiment, computing environment 1700 may include more than one NVMe device 1746, such as a first NVMe device located in DPU 1740 and a second NVMe device located on server 1702 and directly associated with PCIe switch 1720. In another embodiment, DPU 1740 may not include DDR memory 1744 and may include compute storage service (CSS) in place of or as a supplement to DDR memory 1744. For example, computing environment 1700 may include DPU compute storage (CS) memory 1706, which is available to DPU 1740 as part of the CSS. Depending on any suitable interface protocol, such as via Ethernet, InfiniBand, Fibre Channel Remote Direct Memory Access (RDMA), etc., network 1704 may be able to interface with DPU CS memory 1706 via NIC 1742.

[0171] The total memory available for data storage in the computing environment 1700 can be expanded by using DPU 1740 on the system nodes. DPU 1740 can access memory pool 1750 already available on server 1702, such as Double Data Rate (DDR) memory, onboard NVMe devices, on-structure NVMe devices, and CS. Memory pool 1750 may include at least one of DDR memory 1744, NVMe 1746, and DPU CS memory 1706. DPU 1740 can also access available memory of other DPUs that are part of pool 1750, and other DPUs can access available memory of DPU 1740, such as pool 1750. This available memory can be accessed and used for data storage without adding computing resources, such as compute nodes, which are necessary for using other solutions. The available pool 1750 accessible to DPU 1740 can be configured to server 1702 to expand the total memory available for data storage, for example, to reduce the data storage load on CPU 1710 or GPU 1730, which can improve the utilization of their memory for processing. For example, during AI training, model states, residual states, activation functions, and checkpoints can be stored on or offloaded from pool 1650 accessible by the DPU 1740.

[0172] Embodiments of this disclosure may include any suitable materials known in the art (e.g., carbon steel, aluminum, polymers, ceramics, etc.), particularly materials with high thermal conductivity. For example, the components described herein may be manufactured by extrusion and / or machining. In this example, a single body with a fixed cross-sectional area may be produced by an extrusion process.

[0173] Many modifications and other embodiments of the disclosure set forth herein will occur to those skilled in the art upon which this disclosure pertains, having benefited from the teachings presented in the foregoing description and the accompanying drawings. Although the drawings illustrate only certain components of the apparatus and systems described herein, it should be understood that various other components (e.g., components of printed circuit boards, transceivers, cables, etc.) may be used in conjunction with the components described herein. Therefore, it should be understood that this disclosure is not limited to the specific embodiments disclosed, and modifications and other embodiments are also intended to be included within the scope of the appended claims.

[0174] It should be understood that any feature described herein may be claimed in combination with any one or more other features described herein, regardless of whether such features are derived from the same embodiment described.

[0175] Specific details are provided in this specification to provide a thorough understanding of the embodiments. However, those skilled in the art will understand that the embodiments can be practiced without these specific details. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail to avoid obscuring the embodiments.

[0176] While exemplary embodiments of this disclosure have been described in detail herein, it should be understood that the inventive concept may be embodied and used in various ways, and the appended claims are intended to be construed as including such variations, unless limited by the prior art.

Claims

1. A system comprising: A substrate supporting one or more processors; as well as A heat transfer plate, the heat transfer plate including one or more cutouts for accommodating and receiving the one or more processors, wherein the heat transfer plate is configured to transfer heat away from the one or more processors.

2. The system of claim 1, wherein the one or more processors include at least one of a graphics processing unit (GPU) and a central processing unit (CPU).

3. The system of claim 1, further comprising: A fluid conduit that delivers heat transfer fluid around the one or more processors.

4. The system of claim 3, wherein the fluid conduit is integrated into the heat transfer plate.

5. The system of claim 4, further comprising: A highly conductive diffuser is disposed on each of the one or more processors to extract heat from the top surface of the one or more processors, wherein each highly conductive diffuser is also thermally coupled to the heat transfer plate to remove heat from each highly conductive diffuser.

6. The system of claim 4, wherein the one or more processors comprise a plurality of processors, and wherein the heat transfer plate is mounted to the substrate using a plurality of fasteners, the number of fasteners being twice the number of processors among the plurality of processors.

7. The system of claim 6, wherein each highly conductive diffuser is secured to the heat transfer plate using one or more additional fasteners.

8. The system of claim 7, wherein the one or more additional fasteners comprise spring screws.

9. The system of claim 3, wherein the fluid conduit is embedded in the heat transfer plate so that the heat transfer plate can operate as a wet heat transfer plate.

10. The system of claim 9, wherein the heat transfer plate is mounted on the same surface of the substrate as the one or more processors, and wherein the heat transfer plate is thermally coupled to the surface of each of the one or more processors.

11. The system of claim 1, further comprising one or more gap pads.

12. The system of claim 1, wherein each of the one or more cuts corresponds to a corresponding processor among the one or more processors with respect to the die.

13. The system of claim 1, wherein the heat transfer plate includes a fluid conduit incorporated therein, the fluid conduit operating as a manifold to control the flow rate of fluid passing through it.

14. The system of claim 13, wherein the manifold includes a parallel manifold that ensures that the flow of the fluid is distributed substantially uniformly across the one or more processors.

15. The system of claim 1, wherein the substrate is part of a base plate.

16. The system of claim 1, wherein the substrate is part of a PCIe card.

17. A system comprising: substrate; Multiple processors are mounted on the substrate; as well as A heat transfer plate comprising a plurality of slits, each of the plurality of slits corresponding to a different processor among the plurality of processors with respect to a wick, wherein the heat transfer plate includes fluid conduits therein that deliver coolant around the plurality of processors to extract heat generated by the plurality of processors.

18. The system of claim 17, wherein the fluid conduit includes an inlet for receiving the coolant and an outlet for discharging the coolant.

19. The system of claim 17, further comprising: A plurality of highly conductive diffusers, each of the plurality of highly conductive diffusers being disposed above a corresponding processor in the plurality of processors, and each of the plurality of highly conductive diffusers being fastened to the top surface of the heat transfer plate to thermally couple the top surface of the processor to the heat transfer plate. Multiple gap pads, each of the multiple gap pads being disposed between a high-conductivity diffuser and the heat transfer plate in the multiple high-conductivity diffusers; as well as A thermal interface material (TIM) is disposed between the top surface of the processor and the bottom surface of the corresponding highly conductive diffuser, wherein the TIM comprises at least one of the following: thermal grease, thermal tape, thermal pad, gel, adhesive, gap filler, and phase change material.

20. A system comprising: A base plate that supports multiple computing components; A heat transfer plate is fixed to the base plate, and the heat transfer plate includes multiple cutouts, each cutout corresponding to a different computing component among the multiple computing components with respect to the core. as well as At least one diffuser facilitates heat transfer from the upper surface of at least one of the plurality of computing components to a heat transfer fluid conduit.