Computing system partition generator

Abstract

The present disclosure relates to a computing device and method for generating a graphical representation of one or more portions of a computing system, the computing device may include a computing processor and memory, and is configured to access configuration information identifying partitions of the computing system, the computing system comprising an array of systems-on-wafers (SoWs), each SoW of the array of SoWs comprising an array of dies, and generating a graphical representation of at least a portion of the computing system, the graphical representation identifying the partitions and individual dies of the partitions.

Description

[Technical Field]

[0001] [Cross-Reference to Related Applications] This application claims the benefit of priority to U.S. Provisional Application No. 63 / 378,029, entitled "SYSTEM ON WAFER PARTITION GENERATOR", filed on September 30, 2022, the entire disclosure of which is incorporated herein by reference in its entirety for all purposes.

[0002] The present disclosure generally relates to partitioning and / or generating graphical representations of computing systems. [Background Art]

[0003] Certain computing systems can be used and / or specifically configured for high-performance computing and / or computationally intensive applications such as neural network training, neural network inference, machine learning, artificial intelligence, and complex simulation. In some applications, computing systems can be used to perform neural network training. For example, such neural network training can generate data for autopilot systems for vehicles (e.g., automobiles), other autonomous vehicle functions, or advanced driving assistance system (ADAS) functions.

[0004] In high-performance computing systems, there may be high-density processing dies. To analyze and debug high-density dies, it may be desirable to analyze one or more portions of the high-density dies. In computing systems having a large number of processing dies, there are technical challenges associated with analyzing and debugging the dies and associated computing systems. [Brief Summary of the Invention]

[0005] Each of the technological innovations described in the claims has several aspects, and not just one of them alone embodies the desired attributes. Without limiting the scope of the claims, some notable features of this disclosure are briefly described here.

[0006] One aspect of the present disclosure is a computing device for partitioning and / or generating a graphical representation of a computing system. The computing device comprises a computing processor and memory for storing computer executable instructions, and when executed by the computing processor, an operation is performed, the operation includes accessing configuration information that identifies partitions of the computing system and generating a graphical representation of at least a portion of the computing system. The computing system comprises an array of system-on-wafers (SoWs), and each SoW in the array of SoWs comprises an array of dies. Furthermore, the graphical representation identifies partitions and the individual dies within those partitions.

[0007] In computing devices, graphical representations can provide information related to the functionality of individual dies within a partition. Furthermore, this information can indicate whether each individual die is functional, partially functional, or non-functional.

[0008] In computing devices, configuration information can define the voltage supply level and clock frequency for each partition.

[0009] In computing devices, the operation may further include checking for incorrect configuration of configuration information.

[0010] In computing devices, the operation may further include dynamically generating partitions.

[0011] In computing devices, the operation may further include generating a second graphical representation of the partition die, the second graphical representation of which may indicate errors on one or more nodes of a particular die of the partition.

[0012] A computing device may include a display configured to show a graphical representation.

[0013] Another aspect of the present disclosure is a method for generating a graphical representation of a computing system. The method includes accessing configuration information that identifies partitions of the computing system and generating a graphical representation of at least a portion of the computing system. The computing system comprises an array of system-on-wafers (SoWs), and each SoW in the array of SoWs comprises an array of dies. Furthermore, the graphical representation identifies partitions and the individual dies within those partitions.

[0014] In this method, the graphical representation can provide information related to the functionality of each die in the partition. Furthermore, the information related to the functionality of each die in the partition can indicate whether each individual die is functional, partially functional, or non-functional.

[0015] In this method, the configuration information can define the voltage supply level and clock frequency for each partition.

[0016] In this method, the method may further include checking for incorrect configuration of the configuration information.

[0017] In this method, the method may further include dynamically generating partitions.

[0018] In this method, the method may further include generating a second graphical representation of the partition's die, the second graphical representation indicating errors on one or more nodes of a particular die of the partition.

[0019] In this method, the graphical representation can be further displayed on the screen.

[0020] Another aspect of the present disclosure is a non-temporary computer-readable storage medium. The storage medium includes instructions that, when executed by one or more processors, cause a method to be executed that accesses configuration information identifying partitions of a computing system and generates a graphical representation of at least a portion of the computing system. The computing system comprises an array of system-on-wafers (SoWs), each SoW comprising an array of dies. Furthermore, the graphical representation identifies partitions and the individual dies within those partitions.

[0021] For the purpose of summarizing this disclosure, specific aspects, advantages, and novel features of the technological innovation are described herein. It should be understood that not all of such advantages can necessarily be achieved according to any particular embodiment. Thus, the technological innovation may be embodied or implemented to achieve or optimize one advantage or set of advantages as taught herein, without necessarily achieving other advantages that can be taught or suggested herein. [Brief explanation of the drawing]

[0022] Embodiments of the present disclosure will be described as non-limiting examples with reference to the accompanying drawings.

[0023] [Figure 1] An example of a system tray and associated cabinet according to one embodiment is shown.

[0024] [Figure 2]It is a schematic diagram of an exemplary computing system according to various embodiments disclosed in the present specification.

[0025] [Figure 3] An example of a computing system according to an embodiment disclosed in the present specification is shown.

[0026] [Figure 4A-B] FIG. 4A shows an example of a system on wafer (SoW) including a die array.

[0027] FIG. 4B shows an example of nodes included in a die.

[0028] [Figure 5A] An example of a block diagram of exemplary computing systems and computing devices according to various embodiments of the present disclosure is shown.

[0029] [Figure 5B] An example of a command interface according to an embodiment disclosed in the present specification is shown.

[0030] [Figure 5C] An exemplary graphical representation of a die of a SoW array according to an embodiment disclosed in the present specification is shown.

[0031] [Figure 5D] An example of a command interface provided for generating partitions of a SoW array according to an embodiment disclosed in the present specification is shown.

[0032] [Figure 5E] An example of a graphical representation of a 3×2 array of SoW according to an embodiment disclosed in the present specification is shown.

[0033] [Figure 5F] An exemplary graphical representation of nodes of a partition of a 2×2 array of dies according to an embodiment disclosed in the present specification is shown.

[0034] [Figure 5G] A graphical representation of a 2x2 array instance of a single SoW according to embodiments disclosed herein is shown.

[0035] [Figure 6] This specification shows one embodiment of the architecture of a computing device according to several embodiments disclosed herein. [Modes for carrying out the invention]

[0036] The following detailed description of a particular embodiment presents various descriptions of that particular embodiment. However, the technological innovations described herein can be embodied in numerous different ways, for example, as defined and encompassed by the claims. In this description, similar reference numbers and / or terms refer to drawings in which identical or functionally similar elements may be shown. It will be understood that the elements shown in the drawings are not necessarily drawn to scale. Furthermore, it will be understood that a particular embodiment may include more elements and / or subsets of elements shown in the drawings than those shown. In addition, some embodiments may incorporate any suitable combination of features from two or more drawings.

[0037] This disclosure relates to partition generators for computing systems and the use of partition generators, such as generating graphical representations of computing systems, dynamic partitioning, and testing a single system-on-wafer using the partition generator. For example, aspects of this disclosure provide a system for automatically partitioning one or more systems-on-wafers (SoWs). In some aspects, partitions can be defined at various scales or hierarchical levels so as to include a portion of an SoW, a single SoW, a die of two or more SoWs, or multiple SoWs. The systems disclosed herein can identify a hierarchical representation of a computing system by identifying the SoW arrangement and die arrangement of each SoW. The system can also generate partitions based on identifying one or more of these arrangements. In some aspects, one or more of the partitions can be graphically represented to visualize such partitions(s) in the context of various hierarchical representations. Graphical representations can be useful for debugging computing systems.

[0038] In various embodiments of this disclosure, a computing system may comprise one or more computing tiles, each computing tile may comprise a system-on-wafer (SoW) and may be configured to perform computing tasks of the computing system. According to some embodiments, each SoW may comprise multiple dies. For example, each SoW may comprise an array of integrated circuit dies (hereinafter "dies"). Using SoWs, a computing system can achieve high computation density. An SoW may include an integrated cooling system. A system tray may include an array of SoWs supported by a common structure and connected to one another. System trays may be placed in a computing cabinet. SoWs in adjacent computing cabinets may be connected to one another within the computing system. Any appropriate number of SoWs and dies within each SoW may be used in accordance with any appropriate principles and advantages disclosed herein.

[0039] As the demand for computing resources in computing systems increases, high-density computing systems are desired. As mentioned above, a particular computing system may have one or more Systems of Work (SoWs), each SoW may have an array of dies. During computing operation or before installing the SoW into the computing system, the performance of the SoW can be monitored and / or tested to ensure that its performance meets the design specifications. However, monitoring the operation and / or performance of such computing systems can be difficult. For example, the performance of one or more dies in the SoW may degrade, and this degradation in the performance of one or more dies can lead to an overall degradation in the computing system's performance.

[0040] Traditionally, monitoring the performance of large-scale computing systems has involved extensive computing resources. Identifying specific parts of such computing systems experiencing performance and / or errors, and performing post-processing (e.g., debugging) to ensure performance meets its desired specifications, involves considerable computing resources. For example, in a system with six SoWs, each with a 5x5 array of dies, a conventional system could analyze all 150 dies in the computing system to evaluate performance. Furthermore, conventional systems may not be able to easily detect one or more specific dies causing performance degradation in such systems in real time. For example, a conventional system can acquire system logs, store the acquired logs, and analyze the logs. These shortcomings of conventional systems mean that if one or more dies in its SoW become inoperable or experience performance degradation, it can lead to a performance degradation of the entire computing system. Moreover, if a computing system is determined to be experiencing performance degradation, determining which die in the SoW caused the degradation by analyzing the entire computing system, such as logging all computation logs, can lead to inefficient use of computing resources within the computing system.

[0041] To address at least some of the technical challenges described herein, one or more aspects of this disclosure relate to a computing device capable of performing compartmentalization of a computing system at various levels of the system and generating a graphical representation of the computing system. Exemplaryly, the computing device disclosed herein can be communicatively coupled to a computing system and can compartmentalize the Systems of Work (SoW) contained within the computing system. The computing device can then analyze the compartmentalized SoW (e.g., dies contained within the compartmentalized SoW) to identify performance metrics for the dies and, if it identifies that the performance metrics do not meet a particular computing performance specification, it can debug its operation. This disclosure is not limited to the specific computing performance specifications disclosed herein and may be determined based on a particular application.

[0042] As disclosed herein, a System of Work (SoW) of a computing system can be segmented, and the performance of the individual dies contained within the segmented SoW can be monitored and analyzed to determine whether these dies are providing the desired (or designed) computing performance. For example, a single SoW may include 25 dies, each configured to provide its own performance data. For example, the performance metric for each die can be measured based on the telemetry data provided by each die. Thus, the performance of each SoW can be determined based on the measured performance metric for each die. More specifically, each die may include one or more nodes designed to measure the performance metrics of their neighboring nodes. Thus, the performance of the SoW is determined based on the measured performance metrics from the relevant dies.

[0043] In some aspects of this disclosure, after a SoW has been segmented, the segmented SoW may include one or more dies whose performance can be monitored. The process of segmenting a SoW and analyzing the dies within the segmented SoW may be advantageous because it can enable the efficient use of computing resources when investigating the cause of a computing system performance degradation or failure. For example, if a computing system performance degrades, the computing device described herein can segment sections of the SoW and identify the factors leading to the performance degradation. This may be more efficient than conventional systems that involve analyzing the entire SoW.

[0044] Furthermore, when new Systems of Work (SoWs) are added to a computing system, the computing device can monitor the performance metrics of these newly added SoWs by categorizing them. Thus, the systems and methods disclosed herein can enable efficient use of computing resources when analyzing the operational metrics of a computing system and identifying any dies that cause performance degradation. Performance metrics in this disclosure may include, but are not limited to, power consumption, utilization, availability, throughput, and computational latency.

[0045] While embodiments disclosed herein may relate to computing systems having a SoW, any suitable principles and advantages disclosed herein can be applied to computing systems comprising multiple dies partitioned for performing computing tasks.

[0046] The principles and benefits disclosed herein can be applied to any suitable computing device. While aspects of this disclosure are described with respect to exemplary computing components and interactions, those skilled in the art will understand that one or more aspects of this disclosure may be implemented according to various environments, system architectures, computing device architectures, etc. Furthermore, examples are intended to be illustrative and not limiting.

[0047] In non-limiting examples, multiple SoWs can be implemented as computing resources in a computing system. For example, as shown in Figure 1, multiple SoWs can be implemented in one or more cabinets, such as cabinet 120 shown in Figure 1.

[0048] Figure 1 shows an example of a system tray 100 according to one embodiment. As shown, the system tray 100 may include an array of computing tiles 102 connected to each other and supported by structural busbars 104. The structural busbars 104 provide structural support and can supply power to the computing tiles 102 placed thereon. In certain embodiments, each computing tile 102 includes a SoW which includes an array of dies integrated with a cooling solution (e.g., a cold plate). The computing tiles 102 may be called training tiles in neural network training applications. Any suitable number of computing tiles 102 can be connected to each other on the system tray 100. For example, Figure 1 shows six computing tiles 102 connected to each other. The system tray 100 may include an in-tray signaling cable 108 to facilitate communication between each computing tile 102 and an external connection hub (not shown). Each computing tile 102 may include a SoW.

[0049] The computing tiles 102 can be positioned close to each other so that the connections between them are relatively short, such as those established via the in-tray signal transmission cable 108, in order to facilitate high-speed connectivity. The system tray 100 can operate at relatively high power while maintaining mechanical integrity and dissipating enough heat to operate at an appropriate temperature. The illustrated system tray 100 can support high-density integration. For example, the system tray 100 can support a considerable mass while maintaining a relatively low height.

[0050] As further shown in Figure 1, the system tray 100 may have stepped edges 106 positioned along the length of opposing sides of the system tray 100. The stepped edges 106 facilitate sliding the system tray 100 in and out of a cabinet, such as a computing cabinet 120. The system tray 100 can move in and out of the cabinet 120 via slots 122. The system tray 100 may have handles 110 that help facilitate inserting and removing the system tray 100 from the cabinet. The structural busbars 104 may have multiple layers that provide structural and electrical support to the computing tiles 102. While Figure 1 illustrates a specific structure of the system tray 100 and cabinet 120, any suitable principles and benefits disclosed herein can be applied to any suitable computing system.

[0051] In some embodiments, the computing tile 102 may include one or more of a cooling system, a voltage regulator module, a frame structure, a SoW, and a heat dissipation structure. An example of the computing tile 102 is disclosed in International PCT Application No. PCT / US2022 / 040420, entitled "CONNECTOR SYSTEM FOR CONNECTING PROCESSOR SYSTEMS AND RELATED METHODS," which is incorporated herein by reference in its entirety. In certain applications, the computing tile 102 may include a communication interface for communicating data with a computing device. For example, as disclosed herein, a computing device may receive data from one or more SoWs included in the computing tile 102 and perform segmentation of the SoW according to embodiments disclosed herein. Furthermore, the computing device may analyze the performance of the segmented SoW (e.g., dies included in the segmented SoW) and generate a graphical representation of the analyzed performance of the dies and / or SoW. In some applications, the cabinet 120 includes a communication interface and a controller for communicating with the computing device. This application is not limited to the number of Systems of Work (SoWs) that are communicatively coupled to a computing device.

[0052] Figure 2 shows an example of a computing system 200 according to embodiments disclosed herein. The computing system 200 can be implemented in several computing cabinets and can function as a single computing system that can be partitioned to perform various computing tasks. As shown in Figure 2, the computing system 200 may comprise several computing tiles 102. In certain scenarios, these computing tiles 102 can be linked with peripheral components and implemented as part of the computing system 200. Peripheral components may include, but are not limited to, a host 202 and an interface processor such as a network interface processor (NIP) 206.

[0053] In various scenarios, host 202 can be used to identify host addresses associated with one or more computing tiles 102. For example, host 202 in group 210 can provide specific host addresses to computing tiles 102 in group 210, allowing these computing tiles 102 to share the host addresses provided by host 202.

[0054] In some scenarios, the NIP204 can facilitate communication between computing tiles 102. For example, computing tile 102 can exchange data with neighboring computing tiles 102. In other scenarios, computing tile 102 can communicate data with computing devices via the NIP204. The computing device can, for example, receive configuration information for each computing tile 102 via the NIP204 and the host 202.

[0055] In some scenarios, the dies within each computing tile 102 can provide data related to their performance via the NIP 204. For example, the dies within computing tile 102 can send data indicating the availability of their computing node to a computing device via either the NIP 204 or the host 202.

[0056] In certain embodiments, peripheral components may also include memory such as high-bandwidth memory (HBM) 206. In various non-limiting examples, the system tray 100 (shown in Figure 1) may include six computing tiles 102 of group 220. Group 220 may also include respective peripheral components including NIP 204, HBM 206, and host 202.

[0057] Figure 3 shows an example of a computing system 300 according to embodiments disclosed herein. As shown in Figure 3, the computing system 300 may include a SoW array 310. Each SoW 312 of the SoW array 310 may be included, for example, in each computing tile 102 in Figure 1. The SoW array 310 may contain SoW 312 on one or more system trays and / or in one or more cabinets. The SoW array 310 can be used as a computing resource for the computing system 300. The number of SoW 312 in the SoW array 310 can be determined based on a specific application.

[0058] Each SoW312 may include a die array 320, as shown in Figure 3. The die array 320 may include dies 322, for example, as further shown in Figure 3. As shown in Figure 3, each die 322 may be located within the die array 320 and may be logically represented by an array address such as U00-U44. In some embodiments, the interface processor 324 may communicate with the NIP 204 in Figure 2. The interface processor 324 may receive information from the host 202 in Figure 2 via the NIP 204. The interface processor 324 may communicate with the NIP using the Ethernet protocol. In various examples, any appropriate number of dies 322 can be included in the SoW312.

[0059] Figure 4A shows an example of a SoW312 including a die array 320 containing dies 322. Each die 322 can be an integrated circuit die. The dies 322 can be mounted on a SoW312 packaged in a wafer-level packaging structure.

[0060] As shown in Figure 4B, in some embodiments, the die 322 may comprise an array of nodes. The array of nodes may include compute nodes 406 and global nodes 408. In some embodiments, the compute node 406 may include circuitry for performing processing tasks. The global node 408 may generate telemetry data for the die 322. The global node 408 does not have to include circuitry for performing processing tasks. For example, the global node 408 may include pressure, voltage, and temperature (PVT) sensors for monitoring the operating status of the die 322. In some implementations, both the compute node 406 and the global node 408 may include communication interfaces that enable communication with neighboring nodes. For example, each global node 408 may monitor the operating voltage of surrounding nodes by receiving current supply voltages from neighboring nodes via the communication interface. In some implementations, the communication interface for the compute node 406 may be the same as the communication interface for the global node 408.

[0061] In some embodiments, each die 322 can provide performance data related to each node 406. The performance data may refer to information generated from each die. The performance data may include, but is not limited to, environmental information such as the ambient temperature and / or operating temperature of the die(s), operating parameters such as the power supply to each die(s), current and / or voltage measurements of the die(s), and performance information such as usage, bandwidth, and / or latency of the die(s). The performance data may also include data relating to the functionality of parts of the die 322, such as each compute node 406. Each die 322 can be configured to communicate data from the SoW 312 via an input / output interface 412. The input / output interface 412 can also be connected to an interface processor 324 (shown in Figure 3). In this example, the global node 408 can be configured to provide its data to the SoW 312 via interface 412. In some scenarios, the global node 408 can continuously monitor the operating parameters of the compute nodes 406 by enabling a communication interface with the adjacent compute nodes 406. Furthermore, the global node 408 can continuously measure the operating temperature of the die 322.

[0062] Figure 5A shows an example block diagram of a computing system 500 and a computing device 510 according to various embodiments of the present disclosure. The computing system 500 may correspond to computing systems 200 and / or 300, for example, as shown in Figures 2 and 3, respectively. For example, the computing system 500 may comprise at least a SoW array 310 including a SoW 312. In some embodiments, the computing system 500 may be communicatively coupled to a computing device 510. The SoW 312 may communicate with the computing device 510 via a NIPS and a host. In some cases, the computing system 500 and the computing device 510 are connected via a network which may be wireless or wired. This network may include one or more direct communication channels, local area networks, wide area networks, personal area networks, and / or any combination of wired and / or wireless networks such as the Internet. The network may include a specific type of data bus used to transmit data.

[0063] In some embodiments, the computing device 510 may include any suitable computing device(s), such as one or more server computers or one or more desktop computers. In some embodiments, the computing device 510 may store instructions and execute instructions to perform one or more operations of the embodiments disclosed herein. In various embodiments, a user (e.g., a system operator, administrator, developer, etc.) may interact with the computing system 500 by utilizing the computing device 510. In some embodiments, such interaction may be achieved through an interactive graphical user interface, a command line, and / or any other suitable means. For example, the graphical user interface of the computing device 510 may display a graphical representation of the results of processing data received from the computing system 500, such as data relating to the performance of each SoW312 and / or the dies contained in the SoW312. Furthermore, the computing device 510 may provide an interface for providing commands for partitioning SoWs. For example, a user may logically partition SoWs by generating one or more commands specifying partitioning information.

[0064] In some applications, the computing system 500 may include one or more controllers (not shown in Figure 5A) for providing data to the computing device 510. For example, one or more controllers may be implemented within the cabinet 120 in Figure 1, providing data such as configuration information for the computing tiles 102 in Figure 1 within the cabinet 120, performance data for each computing tile 102, and performance data for the dies of the computing tiles 102. Furthermore, the host 202 in Figure 2 may include a controller, which can provide the computing device 510 with configuration information for the computing tiles 102 in the system tray 100. In some embodiments, the controller may receive instructions from the computing device 510 and process data based on the received instructions. For example, the computing device 510 may perform SoW segmentation and request data such as performance metrics for the segmented SoW. In this example, the controller may identify the segmented SoW, process performance metrics such as die(s), bandwidth, and / or latency, and send the processed performance metrics to the computing device 510.

[0065] The computing device 510 can also be configured to manage the configuration of the computing tiles 102. More specifically, the computing device 510 can manage the configuration by identifying each computing tile 102 based on at least the identifier of the computing tile 102, the location of the computing tile 102, and the configuration information of the computing tile 102.

[0066] Figure 5B shows an example of a command interface 530 according to embodiments disclosed herein. The information shown in Figure 5B allows for a concise definition of partitions in a computing system. The command interface 530 can provide various fields for identifying the configuration of the SoW array 310 and the configuration of the dies 322 contained in each SoW 312. As shown in Figure 5B, the command interface 530 may include a SoW array configuration field 532, a definition field 534, a SoW configuration field 536, and an address field 538. The information in these fields together can define partitions in a computing system. The fields in the command interface 530 can be based on a two-dimensional coordinate system and can be specified based on a start coordinate (e.g., top left) and an end coordinate (e.g., bottom right). The partition generator can recognize the hardware to select partitions based on these coordinates. The partition generator can apply specific boot time configuration values, such as the power supply voltage and clock frequency of individual dies. As shown in Figure 5B, the fields are described merely as examples, and more or fewer fields can be used based on specific applications.

[0067] The SoW array configuration field 532 can provide an SoW array configuration. For example, as shown in Figure 5B, the SoWs are arranged in an array (e.g., six SoWs defined as SoW0 to SoW5) as shown in field 532. This arrangement can correspond to the arrangement of the computing tiles 102 contained in the system tray 100 in Figure 1.

[0068] Definition field 534 can be used to define the configuration information for each SoW. For example, definition field 534 in Figure 5B includes the name, location, IP address, and host address of each SoW. Therefore, each SoW can be defined using such field information.

[0069] The SoW configuration field 536 can define configuration information for each SoW. For example, the name is an SoW identification number that can be assigned to each SoW. However, this representation is provided merely as an example, and any other suitable identification number can be used in accordance with any suitable principles and advantages disclosed herein. The location in this particular example in Figure 5B can be defined based on the cabinet 120. For example, SoW0 is contained in the system tray 100 of cabinet 120, and the location field can identify the cabinet 120 containing SoW0. The SoW configuration field 536 can further provide interface processing information for each die 322 of the SoW. For example, as shown in Figure 5B, the dies 322 (U04, U03, U02, U01, U00) of SoW0 are identified by the hostname and channel number of the NIP204 in Figure 2. For example, as described in Figure 2, the host 202 can provide the hostname of each corresponding SoW, and the NIP204 can provide the specific channel number to which each die of the SoW is connected.

[0070] The address field 538 can further provide specific host addresses and names at a higher level corresponding to the group of die 322 (U04, U03, U02, U01, U00).

[0071] The partition generator operating on device 510 can manage incorrect configurations. The partition generator can abstract certain information that may be derived from the information provided by the command interface 530, and still adhere to specific rules for partitioning. For example, the partition generator can prevent the creation of partitions that duplicate hardware resources. This can result from the partition generator being aware of all existing partitions on the hardware system. As another example, the partition generator may not allow IP reuse, as it has checks built to avoid IP collisions. Yet another example is that the system specification schema is free of information duplication and therefore resilient to user input errors. The final system configuration is derived and corrected by construction.

[0072] In some embodiments, the computing device 510 can segment the SoW. Figure 5C shows an example of the SoW arrays and dies included in each SoW according to embodiments disclosed herein. For example, the SoW can be arranged in a 2x3 array (e.g., six SoWs) and mounted in a system tray 100 of a cabinet 120. As shown in Figure 5C, each SoW includes a 5x5 die array 320. In some embodiments, the computing device 510 can logically combine each array of the SoW. For example, the combined SoW may have a 15x10 die array. The number and type of arrays shown in Figure 5C are provided merely as examples, and these numbers and types can be determined based on a specific application.

[0073] In various embodiments, the computing device 510 may provide a command interface 550 for performing partitioning of the SoW, for example, as shown in Figure 5C. As illustrated in Figure 5D, the command interface 550 may define partitions of the SoW, such as partitions 552 and 556. Partition 552 provides partition start and end information, as shown in Figure 5D, and partition 554, as shown in Figure 5C. Partition 554, shown in Figure 5C, corresponds to the partition start and end information shown in partition 556 in Figure 5D. In various embodiments, the computing device 510 provides specific operating parameters for each partition. Operating parameters may include, but are not limited to, supply voltage (Vdd), clock frequency, clock, routing information, and interface information. In some embodiments, these partitions can operate with these operating parameters when the computing system is started up. In some applications, the partitioned SoW can be used for debugging and testing purposes before being implemented in the computing system. For example, partition 554 can be tested with various operating parameters before the system tray 100 is placed in the cabinet 120.

[0074] In some embodiments, the computing device 510 can generate a graphical representation of a die that represents the die's current operating state. The operating state of a die may include, but is not limited to, functional dies (e.g., functional die 560 in Figure 5E), non-functional dies (e.g., non-functional die 566 in Figure 5E), and partially functional dies (e.g., partially functional die 570 in Figure 5E). For example, a functional die may represent a die capable of performing computations, so that these dies can be used to perform computation tasks as part of a computing system. A non-functional die 566 may represent a die that cannot perform computation tasks or system routing. Furthermore, a non-functional die 566 may represent a die with an error (e.g., a functional error) or a die that has been disabled. A partially functional die 570 may represent a die capable of performing a subset of the functions of a functional die 560. For example, a partially functional die 570 may perform signal routing functions and not computation functions in a particular application.

[0075] Figure 5E shows an example of a graphical representation of a 3x2 array SoW (e.g., SoW0-5). The graphical representation in Figure 5E identifies different partitions, each having a specific shading for each die of a particular partition. The graphical representation can be generated based on access to configuration information and other computing system information that defines the partition(s). The configuration information may include any appropriate information shown in Figures 5B and / or 5D.

[0076] As shown in Figure 5E, each SoW can be identified by a specific address, such as the host Internet Protocol address 564, which corresponds to each SoW. The computing device 510 can generate a graphical representation of the current operating state of each die. For example, as shown in Figure 5E, a functional die 560 can be represented by a specific marker, color, etc. Furthermore, in this example, a non-functional die 566 can be represented by a unique marker, such as the X shown on the non-functional die 566 in Figure 5E. In addition, a partially functional die 570 can be represented by annotations such as die 570, as shown in Figure 5E. In some cases, a partially functional die can be used to route data between other dies and cannot be used to compute functions.

[0077] In various applications, users of a computing system (e.g., system engineers, operators, administrators, etc.) can partition the SoW based on these graphical representations. For example, after identifying an unavailable die 566, the computing device 510 can partition the SoW using the command interface 550. For example, a partition input such as partition start:[0,0], partition end:[7,4] may provide partition 562. In another example, a partition input such as partition start:[0,5], partition end:[4,9] may provide partition 568. Thus, these partitions 562,568 can contain only available dies 560. Furthermore, the graphical representation shown in Figure 5E can facilitate debugging, SoW assembly, or computing system maintenance processes. Throughout this disclosure, each die in an array of dies is represented in the form of [y coordinate, x coordinate]. For example, die 592 shown in Figure 5C may be represented as [9,0]. However, this representation is provided merely as an example, and any other suitable index can be used in accordance with any suitable principles and advantages disclosed herein.

[0078] The computing device 510 can also generate partition views of the SoW. These views may include details of the functionality of the die nodes. In some applications, the computing device 510 can generate partition views that include details of individual dies. Figure 5F shows an example of a partition view of a 2x2 die 574, as shown in Figure 5C. For example, the computing device 510 can partition the SoW via the command interface 550 or other user input, using command instructions such as "Partition Start:[1,5], Partition End:[2,6]" corresponding to partition 574, as shown in Figure 5C. Exemplarily, after partitioning, the computing device 510 can generate partition views as shown in Figure 5F. In some embodiments, the computing device 510 can generate queries to receive performance metrics for each die. Die performance metrics may include, for example, their availability, available bandwidth, power, voltage, current, and temperature. Upon receiving performance metrics, the computing device can mark each die based on the received performance metrics. For example, the computing device 510 can determine whether a performance metric is below its threshold for each node of a die. Alternatively or additionally, the computing device 510 can determine whether any node of the die is faulty. If the computing device 510 determines that the performance metric is below its threshold or that there is a fault, it can plot each die with a mark such as color, pattern, or any similar visual representation. For example, as shown in Figure 5F, dies where the performance metric is below its threshold or faulty nodes are represented by different colors or shading 576.

[0079] The graphical representation shown in Figure 5F may be useful in determining where a fault lies within the computing system. In some embodiments, the computing device 510 can generate the graphical representation of Figure 5F for use when debugging errors within a partition. For example, as shown in Figure 5F, an error (e.g., performance degradation or failure) originates from interface portion 572 and spreads to other nodes on the same die and other dies. The error can be identified by an engineer with knowledge of the computing system.

[0080] The graphical representation shown in Figure 5F allows for faster and / or easier error identification than parsing log files. Even when log files are parsed for known patterns, the graphical representation in Figure 5F is useful for identifying one or more unknown failure patterns. This graphical representation enables significantly faster debugging and / or root cause analysis.

[0081] The graphical representation pattern in Figure 5F can indicate a failure in interface 572. Data can be rerouted around interface 572 to operate the relevant die without errors. In some applications, various error patterns of the SoW can be stored in the storage medium of the computing device 510. These stored patterns can be used to determine specific errors associated with a particular pattern. For example, once a specific error pattern is determined, the computing device 510 can compare the determined error pattern with stored error patterns to facilitate the debugging process. While Figure 5F shows a 2x2 array of dies, any suitable principles and benefits of this graphical representation can be applied to any other suitable die array or individual die. Furthermore, the computing device 510 can provide a user interface function to zoom in or zoom out of a particular SoW to represent a detailed view of each die.

[0082] The partitioning disclosed herein can be applied in a variety of useful ways. A dynamic partition generation scheme can be implemented when application software can request various sizes / configurations of partitions at runtime and the partition generation disclosed herein can implement this. A dynamic partition generation scheme can be implemented for system-level testing (SLT) of compute tiles before data center deployment. For example, each compute tile can be partitioned as overlapping 2x2 logical partitions that can be unit-tested. The partition generator can be used to create wrap-around partitions, e.g., 2x2 die partitions containing corner dies of the same SoW. This can be achieved by instantiating the same compute tile four times in the partition generator and then defining a 2x2 die partition for the corner die. Another aspect of the partition generator is that it can also operate with dead hardware annotations when operating in SLT.

[0083] In various embodiments, the computing device 510 can test the SoW and / or system tray before deploying it to a computing system. For example, before a new SoW 582 is deployed to a computing system, the performance and connectivity of the SoW 582 can be tested by utilizing the partitioning configuration of the computing device 510. In some applications, the dies of the SoW 582 (U00, U04, U40, and U44) are used as interface dies for communicating with adjacent SoWs surrounding the SoW 582. In these applications, the computing device 510 can generate logical copies of the SoW 582 to produce four instances of the SoW 582, as shown in Figure 5G. The computing device 510 can then generate a partition 584 containing four corner dies of different instances of the SoW 582. For example, after generating a replicated SoW, the array dimension can be a 10x10 array, and the replicated SoW can be partitioned with commands such as "Partition Start:[4,4], Partition End:[5,5]" corresponding to partition 584 shown in Figure 5G. Partition 584 can provide interface dies (U00, U04, U40, and U44) for SoW582. Therefore, partition 584 can be tested and debugged before deploying SoW582 to a computing system. This is advantageous because the corner dies of SoW582 can be tested without connecting them to other SoWs, and the connectivity of these corner dies can be verified before connecting SoW582 to other SoWs.

[0084] Figure 6 shows one embodiment of the architecture of a computing device 510 according to several embodiments disclosed herein. The general architecture of the computing device 510 shown in Figure 6 may include configurations of computer hardware and software components that can be used to carry out aspects of this disclosure. As shown, the computing device 510 may include a processing unit 612, a network interface 614, a computer-readable media drive 616, and an input / output device interface 618, all of which can communicate with each other via a communication bus. The processing unit 612 may be configured to provide computing resources for executing one or more instructions provided by the memory 620. The network interface 614 may be configured to interact with a computing system as disclosed herein. The network interface 614 may provide a variety of network interfaces, including one or more direct communication channels, a local area network, a wide area network, a personal area network, and / or any combination of wired and / or wireless networks such as the Internet. The network may also be a specific type of data bus used to transmit data. The computer-readable media drive 616 may provide a storage medium according to one or more embodiments disclosed herein. For example, the computer-readable media drive 616 can store various SoW configurations and die configurations. In another example, the computer-readable media drive 616 can also store various error patterns that occurred in the SoW (e.g., dies included in the SoW), as well as any debugging methods and corrective actions applicable to these various error patterns. The input / output device interface 618 can provide interfaces to various computing components such as input components (e.g., keyboard and mouse) and output components (e.g., monitor).

[0085] Memory 620 may include computer program instructions that the processing unit 612 executes to implement one or more embodiments. Memory 620 generally includes RAM, ROM, or other persistent or non-temporary memory. Memory 620 may store an operating system 624 that provides computer program instructions used by the processing unit 612 in the general management and operation of the computing device 510. Memory 620 may further include computer program instructions and other information for implementing embodiments of the present disclosure. For example, in one embodiment, memory 620 may include interface software 622 for communicating with other components.

[0086] The memory may include partitioning instructions 626. Partitioning instructions 626 can be used to partition (e.g., logical partition) an SoW. In some applications, partitioning instructions 626 can be used to manage the configuration of computing tiles 102. More specifically, the computing device 510 can manage the configuration by providing instructions via partitioning instructions 626 to identify each computing tile based on at least the identifier of the computing tile 102, the location of the computing tile 102, and the configuration information of the computing tile 102. As illustrated in Figure 5B, the command interface 530 can provide various fields for identifying the configuration of the SoW array 310 and the configuration of the dies 322 contained in each SoW 312. Furthermore, the command interface 530 may include an SoW array configuration field 532, a definition field 534, an SoW configuration field 536, and an address field 538. The SoW array configuration field 532 can provide the SoW array configuration. For example, as shown in field 532 of Figure 5B, the SoWs are arranged in an array (e.g., six SoWs defined as SoW0-5). This arrangement can correspond to the arrangement of computing tiles 102 contained in the system tray 100 (shown in Figure 1). Definition fields 534 can be used to define the configuration information for each SoW. For example, the definition field 534 in Figure 5B includes the name, location, IP address, and host address of each SoW. Thus, each SoW can be defined using this field information. SoW configuration fields 536 can define the configuration information for each SoW. For example, the name is an SoW identification number that can be assigned to each SoW. However, this identification number is provided merely as an example, and any other suitable type of identifier can be used in accordance with any suitable principles and advantages disclosed herein. The location in this particular example in Figure 5B can be defined based on the cabinet 120.For example, SoW0 is contained in the system tray 100 of cabinet 120, and the location field can identify cabinet 120 containing SoW0. The SoW configuration field 536 can further provide interface processing information for each 322 of the SoW. For example, as shown in Figure 5B, the dies 322 of SoW0 (U04, U03, U02, U01, U00) are identified by the hostname and channel number of the NIP204 (shown in Figure 2). For example, as described in Figure 2, host 202 can provide the hostname of each corresponding SoW, and NIP204 can provide the specific channel number to which each die of the SoW is connected. The address field 538 can further provide a specific higher-level host address and name corresponding to the group of dies (U04, U03, U02, U01, U00). The fields shown in Figure 5B are illustrated as examples only, and more or fewer fields can be used based on specific applications.

[0087] In some embodiments, the computing device 510 can partition a SoW by utilizing a partitioning instruction 626. Figure 5C shows an example of the SoW array and dies contained in each SoW. For example, the SoWs can be arranged in a 2x3 array (e.g., 6 SoWs) and mounted in the system tray 100 of the cabinet 120. As shown in Figure 5C, each SoW can include a 5x5 die array 320. In some embodiments, the computing device 510 can provide instructions to logically combine each array of the SoWs. For example, the combined SoW may have a 15x10 array of dies. The number and type of arrays shown in Figure 5C are provided as examples only, and these numbers and types can be determined based on a specific application.

[0088] In various embodiments, the partitioning instruction 626 may also be provided via a command interface 550 for performing partitioning of the SoW, for example, as shown in Figure 5C. As illustrated in Figure 5D, the command interface 550 may define partitions of the SoW, such as partitions 552 and 556. Partition 552 provides partition start and end information, as shown in Figure 5D, and partition 554, as shown in Figure 5C. Partition 554, shown in Figure 5C, corresponds to the partition start and end information shown in partition 556 in Figure 5D. In various embodiments, the computing device 510 provides specific operating parameters for each partition. Operating parameters may include, but are not limited to, supply voltage (Vdd), clock frequency, clock, routing information, interface information, etc. In some embodiments, these partitions may operate with these operating parameters when the computing system is started up. In some applications, the partitioned SoW can be used for debugging and testing purposes before being implemented in the computing system. For example, partition 554 can be tested with various operating parameters before the system tray 100 is placed in the cabinet 120.

[0089] Memory 620 may also contain graphical representation instructions 628. By utilizing the graphical representation instructions 628, the computing device 510 can provide instructions to the processing unit 612 for generating a partition view of the SoW.

[0090] In some embodiments, the computing device 510 can execute graphical representation instructions to generate a graphical representation of a die that represents the die's current operating state. This may include accessing configuration information that defines partitions. It may also access additional computing system information regarding the die's operating state and / or performance. The operating state of a die may include, but is not limited to, functional dies, non-functional dies, and partially functional dies. For example, a functional die may represent a die capable of performing computations, so that these dies can be used to perform computation tasks as part of a computing system. A non-functional die may represent a die that cannot perform computation tasks, such as a die that does not meet certain specifications for performing a task or computation. Furthermore, a non-functional die may represent a die with errors (e.g., functional errors) or a die that has been disabled. A partially functional die may represent a die capable of performing a subset of the functions of a functional die, such as certain functions other than computation. For example, certain functions may include, but are not limited to, routing functions and interface functions. As shown in Figure 5E above, each SoW may be identified by a specific address, such as the host Internet Protocol address 564 corresponding to each SoW. The computing device 510 can generate a graphical representation of the current operating state of each die. For example, as shown in Figure 5E, a functional die 560 can be represented by a specific marker, color, etc. Furthermore, in this example, a non-functional die 566 can be represented by a unique marker, as shown on the partially functional die 566 in Figure 5E. In addition, dies for limited purposes can be represented by annotations such as die 570, as shown in Figure 5E.

[0091] In various applications, users of the computing system (e.g., system engineers, operators, administrators, etc.) can partition the SoW based on these graphical representations. For example, after identifying an unavailable die 566, the computing device 510 can partition the SoW using the command interface 550. For instance, a partition input such as partition start:[0,0], partition end:[7,4] can provide a partitioned area 562. In another example, a partition input such as partition start:[0,5], partition end:[4,9] can provide a partitioned area 568. Thus, these partitions 562,568 can contain only available dies 560. Furthermore, the graphical representation shown in Figure 5E can facilitate debugging, SoW assembly, or computing system maintenance processes.

[0092] In some applications, the processing unit 612 can execute instructions to generate a partition view at the die level. Figure 5F shows an example of a partition view of a 2x2 die 574, as shown in Figure 5C. For example, the computing device 510 can provide a command interface 550 to partition the SoW (by executing partition instruction 626), such as a command instruction "Partition Start: [1,5], Partition End: [2,6]" corresponding to the partition 574 shown in Figure 5C. Exemplary, after partitioning, the computing device 510 can execute a graphical representation instruction 628 to generate a partition view as shown in Figure 5F. In some embodiments, the computing device 510 can generate queries to receive performance metrics for each die. Die performance metrics may include, for example, their availability, available bandwidth, power, voltage, current, and temperature. Upon receiving performance metrics, the computing device can mark each die based on the received performance metrics. For example, the computing device 510 can determine for each die whether the performance metrics are below their threshold. If the computing device 510 determines that the performance metric is below its threshold, it can plot each die with a mark such as color, pattern, or any similar visual representation. For example, as shown in Figure 5F, dies that meet the performance metric threshold are represented by different colors 576.

[0093] Memory 620 may also include post-processing instructions 630. In some embodiments, processing unit 512 may execute post-processing instructions 630 to debug the die based on the generated partition view. In some embodiments, computing device 510 may generate a graphical representation of Figure 5F for use when debugging errors in a partition by executing post-processing instructions 630. For example, as shown in Figure 5F, an error (e.g., performance degradation or failure) originates from interface portion 572 and spreads to other nodes on the same die and other dies. The error can be identified by an engineer with knowledge of computing systems.

[0094] The graphical representation shown in Figure 5F allows for faster and / or easier error identification than parsing log files. Even when log files are parsed for known patterns, the graphical representation in Figure 5F is useful for identifying one or more unknown failure patterns. This graphical representation enables significantly faster debugging and / or root cause analysis.

[0095] The graphical representation pattern in Figure 5F can indicate a failure in interface 572. Data can be rerouted around interface 572 to operate the relevant die without errors. In some applications, various error patterns of the SoW can be stored in the storage medium of the computing device 510. These stored patterns can be used to determine specific errors associated with a particular pattern. For example, once a specific error pattern is determined, the computing device 510 can compare the determined error pattern with stored error patterns to facilitate the debugging process. While Figure 5F shows a 2x2 array of dies, any suitable principles and benefits of this graphical representation can be applied to any other suitable die array or individual die. Furthermore, the computing device 510 can provide a user interface function to zoom in or zoom out of a particular SoW to represent a detailed view of each die.

[0096] In various embodiments, the computing device 510 can test the SoW and / or system tray before deploying it to a computing system by executing post-processing instructions 630. For example, before a new SoW 582 is deployed to a computing system, the performance and connectivity of the SoW 582 can be tested by utilizing the partitioning configuration of the computing device 510. In some applications, the dies of the SoW 582 (U00, U04, U40, and U44) are used as interface dies for communicating with adjacent SoWs surrounding the SoW 582. In these applications, the computing device 510 can generate logical copies of the SoW 582 to produce four instances of the SoW 582, as shown in Figure 5G. The computing device 510 can then generate a partition 584 containing four corner dies of different instances of the SoW 582. For example, after generating a replicated SoW, the array dimension can be a 10x10 array, and the replicated SoW can be partitioned with commands such as "Partition Start:[4,4], Partition End:[5,5]" corresponding to partition 584 shown in Figure 5G. Partition 584 can provide interface dies (U00, U04, U40, and U44) for SoW582. Therefore, partition 584 can be tested and debugged before deploying SoW582 to a computing system. This is advantageous because the corner dies of SoW582 can be tested without connecting them to other SoWs, and the connectivity of these corner dies can be verified before connecting SoW582 to other SoWs.

[0097] The computing systems disclosed herein can be implemented using a variety of processing systems. Such processing systems can be used and / or specifically configured for high-performance computing and / or computationally intensive applications such as neural network training, neural network inference, machine learning, artificial intelligence, and complex simulations. In some applications, the processing systems can be used to perform neural network training. For example, such neural network training can generate data for vehicles (e.g., automobiles), other autonomous vehicle functions, or autopilot systems for advanced driver-assistance systems (ADAS) functions.

[0098] Unless the context clearly indicates otherwise, terms such as “comprise,” “comprising,” “include,” and “including” throughout the specification and claims should be interpreted in a comprehensive sense, as opposed to an exclusive or exhaustive sense, i.e., “including but not limited to.” The term “combined” as commonly used herein refers to two or more elements that are directly connected or that can be connected by one or more intermediate elements. Similarly, the term “connected” as commonly used herein refers to two or more elements that are directly connected or that can be connected by one or more intermediate elements. Furthermore, the terms “as specified herein,” “above,” “below,” and similar terms, when used in this application, refer to the entire application and not to any particular part thereof. Where the context allows, the terms in the above detailed descriptions that use singular or plural may also include plural or singular, respectively. The term “or” in relation to a list of two or more items encompasses all of the following interpretations of the term, namely any of the items in the list, all of the items in the list, and any combination of the items in the list.

[0099] Furthermore, conditional language used herein, such as "can, could," "might, may," and "e.g., for example, such as," is generally intended to convey that certain embodiments include certain features, elements, and / or states, but other embodiments do not, unless otherwise specified or understood in the context in which they are used. Therefore, such conditional language is generally not intended to imply that features, elements, and / or states are required in any way in one or more embodiments.

[0100] The above description has been written with reference to specific embodiments. However, the above exemplary description is not intended to be exhaustive or to limit the invention to the exact form described. Many modifications and variations are possible in light of the above teachings. Thereafter, those skilled in the art will be able to best utilize the technology and various embodiments with various modifications suitable for various applications.

[0101] While the present disclosure and embodiments have been described with reference to the accompanying drawings, various changes and modifications will be apparent to those skilled in the art. Such changes and modifications should be understood to be within the scope of the present disclosure.

Claims

1. A computing device for generating a graphical representation of a computing system, The system comprises a computing processor and a memory that stores computer executable instructions that, when executed by the computing processor, cause an operation to be performed, and the operation is A step of accessing configuration information that identifies a partition of the computing system, wherein the computing system comprises an array of system-on-wafers (SoWs), each SoW in the array of SoWs comprises an array of dies, the partition is a logical partition, and at least one of the logical partitions comprises (a) a subset of dies from one SoW, or (b) dies from one or more SoWs. A step of generating a graphical representation of at least a part of the computing system, wherein the graphical representation identifies the partition and the individual dies of the partition, Computing devices, including [this].

2. The computing device according to claim 1, wherein the graphical representation provides information relating to the function of the individual dies of the partition.

3. The computing device according to claim 2, wherein the information relating to the function of each die in the partition indicates whether each of the individual dies is functional, partially functional, or non-functional.

4. The computing device according to claim 1, wherein the operation further includes a step of checking for an incorrect configuration of the configuration information.

5. The computing device according to claim 1, wherein the operation further includes the step of dynamically generating the partition.

6. The computing device according to claim 1, wherein the operation further includes the step of generating a second graphical representation of a die of one of the partitions, the second graphical representation indicating an error on one or more nodes of a particular die of the one partition.

7. The computing device according to claim 1, further comprising a display configured to display the graphical representation.

8. A method for generating a graphical representation of a computing system using a computing device having a computing processor, The steps of accessing configuration information that identifies a partition of the computing system using the computing processor, wherein the configuration information is stored in memory, the computing system comprises an array of system-on-wafers (SoWs), each SoW in the array of SoWs comprises an array of dies, the partition is a logical partition, and at least one of the logical partitions comprises (a) a subset of dies from one SoW, or (b) dies from one or more SoWs. A step of generating a graphical representation of at least a part of the computing system using the computing processor, wherein the graphical representation identifies the partition and the individual dies of the partition. Methods that include...

9. The method according to claim 8, wherein the graphical representation provides information relating to the function of the individual dies of the partition.

10. The method according to claim 9, wherein the information relating to the function of each die in the partition indicates whether each of the individual dies is functional, partially functional, or nonfunctional.

11. The method according to claim 8, further comprising the step of checking for an incorrect configuration of the configuration information using the computing processor.

12. The method according to claim 8, further comprising the step of dynamically generating the partition using the computing processor.

13. The method of claim 8, further comprising the step of the computing processor generating a second graphical representation of a die of one of the partitions, wherein the second graphical representation indicates an error on one or more nodes of a particular die of the one partition.

14. The method according to claim 8, further comprising displaying a graphical representation on the display.

15. A non-temporary computer-readable storage medium that, when executed by one or more processors, includes instructions causing the method according to claim 8 to be performed.

Images