A GPU card power supply input system and server

Abstract

The application relates to the field of electronic technology and discloses a GPU card power input system and a server. The GPU card power input system comprises a GPU board, a GPU card, a capacitor board and a first power cable. The GPU board is provided with a fifth power connector. The GPU card is arranged on the GPU board and has a sixth power connector. The capacitor board is provided with a decoupling capacitor. The first power cable comprises a first cable segment and a second cable segment, and the first cable segment and the second cable segment are butted through the capacitor board. The sixth power connector on the GPU card is connected to the fifth power connector on the GPU board through the first power cable. Compared with the prior art, the dynamic noise of the embodiment of the application is smaller, and the sum of total pressure drops is easier to meet the requirements of the GPU card, so that the GPU card can normally and highly perform, and the performance of the whole server is improved.

Description

Technical Field

[0001] This invention relates to the field of electronic technology, and in particular to a GPU (Graphics Processing Unit) card power input system and server. Background Technology

[0002] With increasingly powerful chips, the power consumption of electronic products is inevitably increasing. Therefore, power supply stability design is one of the most important factors for product success. If the power system of an electronic product is unstable and the voltage value does not meet the requirements of the chip or board, it can lead to substandard product performance, or even prevent the operation of critical programs or applications, or even cause the system to crash.

[0003] With the increasing popularity of AI applications, the configuration of multiple high-power GPU cards is a common specification for AI servers. Due to the diversity and complexity of application scenarios, board designers need to ensure that all GPU cards can run at their maximum performance in each application scenario to achieve the overall system specification design goals.

[0004] for example Figure 1 In a certain AI server system (supporting 8 high-power GPU cards), the 12V power input of the high-power GPU cards starts from the PSU (Power Supply Unit), passes through the PDB (Power Distribution Board), and then goes to the power connector Conn1A of the GPU board (referring to the board that mounts the GPU cards). Then, a first power cable is used to connect the power connector Conn1A to the 12V input power connector Conn1B on the GPU card.

[0005] The requirements for 12V power input of high-power GPU cards are shown in the table below:

[0006]

[0007] This means that under any circumstances (including DC voltage drop and dynamic noise), the voltage at the power connector Conn1B must not be lower than 11.04V (single card peak operating current up to 70A), which means the total voltage drop must not be greater than 960mV (based on PSU standard output 12V).

[0008] for Figure 1 In traditional designs, under harsh working conditions, where seven GPU cards operate at peak current, the eighth GPU card (e.g.) may also experience issues. Figure 1The image shown dynamically flips from 0A to 70A (i.e., transitions from idle mode to accelerated computing mode). At this point, the 8th GPU card, besides experiencing a lower DC voltage drop due to the preceding 7 GPUs (because these 8 GPUs share the common power path of "PSU->PDB->GPU board"), also suffers from its own voltage drop during the flip (dynamic noise). Dynamic noise needs to be addressed by decoupling capacitors (which are only effective when very close to the current flip point), however... Figure 1 In this case, all the decoupling capacitors are located near the power connector Conn1A on the GPU board, which is too far from the GPU card (i.e., the current inversion point). They cannot play a significant role, often resulting in the voltage at the power connector Conn1B being lower than the GPU card's 11.04V requirement (i.e., DC voltage drop + dynamic noise > 960mV), causing the GPU card to malfunction. Summary of the Invention

[0009] The purpose of this invention is to provide a GPU card power input system and server to overcome the shortcomings of the prior art, which cause the GPU card to malfunction due to the decoupling capacitor being too far away from the GPU card.

[0010] To achieve this objective, the present invention adopts the following technical solution:

[0011] A GPU card power input system, comprising:

[0012] GPU board, wherein a fifth power connector is provided on the GPU board;

[0013] A GPU card is mounted on a GPU board, and the GPU card has a sixth power connector;

[0014] A capacitor board, wherein a decoupling capacitor is provided on the capacitor board;

[0015] A first power cable, comprising a first cable segment and a second cable segment, wherein the first cable segment and the second cable segment are connected through the capacitor plate.

[0016] The sixth power connector on the GPU card is connected to the fifth power connector on the GPU board via the first power cable.

[0017] Optionally, the system further includes: a power supply unit and a power distribution unit;

[0018] The power supply unit is used to convert standard AC power into low-voltage DC power and has a first power connector.

[0019] The power distribution unit is used to distribute the low-voltage DC power provided by the power supply unit to the GPU board, and has a second power connector and a third power connector that are electrically connected to each other.

[0020] The GPU board also has a fourth power connector, which is electrically connected to each of the fifth power connectors respectively.

[0021] The first power connector is electrically connected to the second power connector, and the third power connector is electrically connected to the fourth power connector.

[0022] Optionally, the first power connector and the second power connector are electrically connected via terminal contacts, and the third power connector and the fourth power connector are electrically connected via a second power cable.

[0023] Optionally, there are multiple fifth power connectors, GPU cards, capacitor boards, and power cables on the GPU board, and the fifth power connector on the GPU board, the sixth power connector on the GPU card, the capacitor board, and the first power cable are in a one-to-one correspondence.

[0024] The sixth power connector of each GPU card is electrically connected to the corresponding fifth power connector on the GPU board via the corresponding first power cable.

[0025] Optionally, one end of the first cable segment is electrically connected to the fifth power connector on the GPU board, and the other end is electrically connected to one end of the second cable segment via a capacitor board. The other end of the second cable segment is electrically connected to the sixth power connector on the GPU card.

[0026] The actual length of the second cable segment does not exceed a preset length threshold, which is the maximum cable length at which the actual dynamic noise generated does not exceed the standard noise threshold.

[0027] Optionally, the length of the first cable segment is greater than the length of the second cable segment.

[0028] Optionally, the GPU board is provided with a PCIE (peripheral component interconnect express, high-speed serial computer expansion bus standard) connector, and the GPU card is inserted into the PCIE connector.

[0029] Optionally, the decoupling capacitors include two 1000uF capacitors and two 470uF capacitors.

[0030] A server comprising a GPU card power input system as described in any of the preceding claims.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] This invention splits the original power cable into two segments: a first segment and a second segment. These segments are connected by a small capacitor bank containing decoupling capacitors to meet the current switching requirements of the GPU card. With this design, the dynamic noise of the GPU card is only affected by the length of its adjacent cable segment (the shorter the cable, the smaller the corresponding inductance), rather than by the length of the entire power cable. Therefore, compared to existing technologies, this invention exhibits lower dynamic noise, and the total voltage drop more easily meets the requirements of the GPU card, ensuring its normal high-performance operation and thus improving the overall performance of the server. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] The structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0035] Figure 1 A schematic diagram of a traditional GPU card power input system;

[0036] Figure 2 This is a schematic diagram of the GPU card power input system provided in an embodiment of the present invention.

[0037] Diagram description: Power supply unit 1, first power connector 11, power distribution unit 2, second power connector 21, third power connector 22, GPU board 3, fourth power connector 31, fifth power connector 32, GPU card 4, sixth power connector 41, capacitor board 5, first power cable 6, first cable segment 61, second cable segment 62, second power cable 7. Detailed Implementation

[0038] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0039] In the description of this invention, it should be understood that the terms "upper," "lower," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. It should be noted that when a component is considered to be "connected" to another component, it can be directly connected to the other component or there may be a component positioned centrally in the connection.

[0040] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0041] It should be noted that the decoupling capacitor and the current reversal point (i.e.) Figure 1 The dynamic noise (AC voltage drop) on the path between the power connectors Conn1B and Conn2B is given by the following formula. The decision is made based on the GPU card's current flip-flop rate (di / dt). Therefore, the dynamic noise (AC voltage drop) between power connectors Conn1A and Conn1B is entirely determined by the inductance of the power cable between them. The inductance of the power cable is primarily related to its length and diameter. Cable length is often constrained by the chassis design and cannot be changed, while cable diameter is a secondary factor and is often limited by the choice of connector.

[0042] Therefore, to solve the problem of excessive dynamic noise and insufficient link voltage drop due to excessively long power cables of GPU cards in traditional solutions, the solution of this invention is: based on the principle of dynamic noise generation... By moving the decoupling capacitor as close as possible to the GPU card, the inductance L in the formula is reduced, thereby reducing dynamic noise and meeting the voltage drop requirements of the GPU card.

[0043] For details, please refer to Figure 2 This invention provides a GPU card power input system, comprising:

[0044] GPU board 3, which has a fifth power connector 32;

[0045] GPU card 4 is mounted on GPU board 3, and GPU card 4 has a sixth power connector 41;

[0046] Capacitor plate 5, which is equipped with decoupling capacitors;

[0047] The first power cable 6 includes a first cable segment 61 and a second cable segment 62, and the first cable segment 61 and the second cable segment 62 are connected through the capacitor plate 5.

[0048] The sixth power connector 41 on the GPU card 4 is connected to the fifth power connector 32 on the GPU board 3 via the first power cable 6.

[0049] In this embodiment of the invention, the original power cable is split into two segments: a first segment 61 (denoted as cable1) and a second segment 62 (denoted as cable2). Cable1 and cable2 are connected by a small capacitor plate 5, on which decoupling capacitors are placed to meet the current switching requirements of the GPU card 4. With this design, the dynamic noise of the GPU card 4 is only affected by the length of cable2 (the shorter the cable length, the smaller the corresponding inductance), rather than by the length of the entire power cable. Therefore, compared with the prior art, the dynamic noise of this embodiment of the invention is lower, and the sum of the total voltage drops is more likely to meet the requirements of the GPU card 4.

[0050] To meet the design requirements, taking the connection of cable2 between GPU card 4 and capacitor board 5 as an example, the actual length of cable2 does not exceed the preset length threshold. The preset length threshold is the maximum length of cable that ensures the actual dynamic noise generated does not exceed the standard noise threshold.

[0051] Understandably, different capacities and / or numbers of decoupling capacitors can be soldered onto capacitor board 5 to suit different GPU cards 4. Meanwhile, the length of cable 2 can be flexibly adjusted within a reasonable range according to different chassis designs.

[0052] To enable power input to the GPU card 4, the GPU card power input system provided in this embodiment of the invention further includes a power supply unit 1 (PSU) and a power distribution unit 2 (PDB).

[0053] The power supply unit 1 is used to convert standard AC power into low-voltage DC power and has a first power connector 11.

[0054] The power distribution unit 2 is used to distribute the low-voltage DC power provided by the PSU to the GPU board 3, and has a second power connector 21 and a third power connector 22 that are electrically connected to each other.

[0055] The GPU board 3 also has a fourth power connector 31, which is electrically connected to each of the fifth power connectors 32.

[0056] The first power connector 11 is electrically connected to the second power connector 21, and the third power connector 22 is electrically connected to the fourth power connector 31.

[0057] In one alternative embodiment, the first power connector 11 and the second power connector 21 are electrically connected via terminal contacts, and the third power connector 22 and the fourth power connector 31 are electrically connected via the second power cable 7.

[0058] Thus, the main power supply path is: PSU -> PDB -> GPU board 3 -> GPU card 4. Because the decoupling capacitors placed close to GPU card 4 effectively reduce the dynamic noise generated by the power cables in the GPU board 3 -> GPU card 4 section, the total voltage drop of GPU card 4 is more likely to meet design requirements, thus ensuring that GPU card 4 can operate normally.

[0059] In practical applications, the number of the fifth power connector 32, GPU card 4, capacitor board 5, and power cable on GPU board 3 can all be multiple, and the fifth power connector 32 on GPU board 3, the sixth power connector 41 on GPU card 4, capacitor board 5, and first power cable 6 are in a one-to-one correspondence.

[0060] The sixth power connector 41 of each GPU card 4 is electrically connected to the corresponding fifth power connector 32 on the GPU board 3 via the corresponding first power cable 6.

[0061] This ensures that in application scenarios with multiple high-power GPU cards, each GPU card can operate normally and at high performance, achieving the overall system specification design goals.

[0062] Below is a comparison example:

[0063] In a certain AI server, system architecture design and... Figure 1Consistent with the standard configuration, it supports eight 350W GPU cards (TDP), each with a peak current of up to 70A. The GPU board has a PCIe connector, into which the GPU cards are inserted. The PSU output voltage is 12V. With seven GPU cards operating at peak current, the DC voltage drop across Conn1A from the PSU to the GPU board is 370mV (including the PDB board voltage drop, the inter-board power cable voltage drop, and the GPU board's voltage drop). Using a conventional GPU power cable design (300mm), the dynamic noise is 670mV. Therefore, the total power link voltage drop is 1040mV, exceeding 960mV and surpassing the GPU card specifications.

[0064] When the GPU power cable solution is changed to the embodiment of this invention (e.g.) Figure 2 As shown, cable1 is 250mm long, cable2 is 50mm long, and capacitor board 5 is equipped with two 1000uF capacitors and two 470uF capacitors. At this time, the dynamic noise is only 560mV, and the total voltage drop of the power link is 930mV, which meets the requirements of GPU card 4 specification.

[0065] The comparison shows that the embodiments of the present invention effectively reduce dynamic noise and reduce the total voltage drop of the power supply link.

[0066] Furthermore, this embodiment of the invention also provides a server including the GPU card power input system described above. Because this GPU card power input system splits the original power cable connecting the GPU board 3 and the GPU card 4 into two, and the two cable segments are connected via a capacitor board 5, the distance between the decoupling capacitor on the capacitor board 5 and the GPU card 4 can be significantly reduced within the limited space of the chassis. This allows the decoupling capacitor to play its actual role, effectively reducing dynamic noise and ensuring that the GPU card 4 can operate normally and at high performance, thereby improving the overall performance of the server.

[0067] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power input system for a graphics processing unit (GPU) card, characterized in that, include: GPU board, wherein a fifth power connector is provided on the GPU board; A GPU card is mounted on a GPU board, and the GPU card has a sixth power connector; A capacitor board, wherein a decoupling capacitor is provided on the capacitor board; A first power cable, comprising a first cable segment and a second cable segment, wherein the first cable segment and the second cable segment are connected through the capacitor plate. The sixth power connector on the GPU card is connected to the fifth power connector on the GPU board via the first power cable. One end of the first cable segment is electrically connected to the fifth power connector on the GPU board, and the other end is electrically connected to one end of the second cable segment via a capacitor board. The other end of the second cable segment is electrically connected to the sixth power connector on the GPU card. The actual length of the second cable segment does not exceed a preset length threshold, which is the maximum cable length that ensures the actual dynamic noise generated does not exceed a standard noise threshold. The length of the first cable segment is greater than the length of the second cable segment.

2. The GPU card power input system according to claim 1, characterized in that, The system also includes: a power supply unit and a power distribution unit; The power supply unit is used to convert standard AC power into low-voltage DC power and has a first power connector. The power distribution unit is used to distribute the low-voltage DC power provided by the power supply unit to the GPU board, and has a second power connector and a third power connector that are electrically connected to each other. The GPU board also has a fourth power connector, which is electrically connected to each of the fifth power connectors respectively. The first power connector is electrically connected to the second power connector, and the third power connector is electrically connected to the fourth power connector.

3. The GPU card power input system according to claim 2, characterized in that, The first power connector and the second power connector are electrically connected via terminal contacts, and the third power connector and the fourth power connector are electrically connected via a second power cable.

4. The GPU card power input system according to claim 1, characterized in that, The number of the fifth power connector, GPU card, capacitor board and first power cable on the GPU board are multiple, and the fifth power connector on the GPU board, the sixth power connector on the GPU card, the capacitor board and the first power cable are in a one-to-one correspondence. The sixth power connector of each GPU card is electrically connected to the corresponding fifth power connector on the GPU board via the corresponding first power cable.

5. The GPU card power input system according to claim 1, characterized in that, The GPU board is equipped with a PCIe standard card connector, and the GPU card is inserted into the PCIe standard card connector.

6. The GPU card power input system according to claim 1, characterized in that, The decoupling capacitors include two 1000uF capacitors and two 470uF capacitors.

7. A server, characterized in that, The server includes the GPU card power input system as described in any one of claims 1 to 6.

Images