A GPU testing tool and a GPU testing system applying the same

Abstract

The application relates to a GPU test tool and a GPU test system applying the test tool, and relates to the technical field of GPU testing; wherein the GPU test tool comprises: a pressing driving structure for applying a preset pressure to a tested GPU, so that the tested GPU is accurately attached to a preset test mainboard; a pressing detection module for detecting the pressing state of the pressing driving structure and outputting a corresponding pressing detection signal; and an execution control module electrically connected with the pressing driving structure and the pressing detection module, respectively, for receiving the pressing detection signal of the pressing detection module and controlling the pressing action of the pressing driving structure based on the pressing detection signal. The application has the effect of optimizing the test performance of the GPU test tool.

Description

Technical Field

[0001] This application relates to the field of GPU testing technology, and in particular to a GPU testing fixture and a GPU testing system using the aforementioned testing fixture. Background Technology

[0002] During GPU production testing and R&D verification, a testing fixture is needed to precisely bond the GPU under test to the test motherboard to ensure a stable electrical connection for subsequent performance testing. The bonding effect directly determines the accuracy of the test data and the integrity of the GPU. Therefore, the core requirement of GPU testing fixtures is to achieve precise and reliable control over the bonding process.

[0003] In existing technologies, traditional GPU testing fixtures often employ a single driving component (such as a cylinder or simple electric actuator) combined with a fixed positioning structure to achieve the pressing action. This relies on preset control of the driving component to regulate the pressing action, making it impossible to determine the actual state during the pressing process (e.g., whether the pressing is complete or whether the applied pressure is sufficient). Furthermore, GPU pins are delicate and fragile; excessive pressing pressure can easily lead to pin deformation and damage, while insufficient pressure or incomplete pressing can cause poor contact between the GPU and the test motherboard, resulting in test signal interruption and data distortion, severely impacting test efficiency and the reliability of test results. Different GPU models have varying thicknesses and pin layouts, requiring different pressing conditions. However, the pressing parameters of traditional testing fixtures (such as the pressing stroke of the driving component) are mostly fixed settings, unable to be dynamically adjusted based on the actual pressing state. They can only be manually adjusted to adapt to different GPU models, which is not only cumbersome and time-consuming but also prone to insufficient pressing accuracy due to manual adjustments, increasing the risk of GPU failure. The risk of damage is a concern. Setting up multiple test fixtures to adapt to different models would incur additional costs. Therefore, existing GPU test fixtures have problems such as not being able to know the pressing state and poor adaptability of the pressing action, which makes it impossible to meet the high-precision and high-reliability testing requirements of GPUs. Therefore, improvements are needed. Summary of the Invention

[0004] To optimize the testing performance of GPU testing fixtures and meet the requirements for high-precision and high-reliability testing of GPUs, this application provides a GPU testing fixture and a GPU testing system using the aforementioned testing fixture.

[0005] Firstly, this application provides a GPU testing fixture, which adopts the following technical solution: The pressure-driven structure is used to apply a preset pressure to the GPU under test, so that the GPU under test is precisely attached to the preset test motherboard; The compression detection module is used to detect the compression state of the compression driving structure and output the corresponding compression detection signal. The execution control module is electrically connected to the pressing drive structure and the pressing detection module, respectively, and is used to receive the pressing detection signal from the pressing detection module and control the pressing action of the pressing drive structure based on the pressing detection signal.

[0006] By adopting the above technical solution and setting up a pressing detection module and an execution control module, the physical state (such as pressure and stroke) during the pressing process can be monitored in real time, and the action of the pressing drive structure can be dynamically adjusted based on the feedback signal. This achieves precise and proactive control of the pressing process, avoiding the problems of excessive pressure damaging the GPU or insufficient pressure causing poor contact caused by fixed parameters in traditional open-loop pressing.

[0007] Preferably, the pressing drive structure includes a drive unit, a pressure plate, and an elastic guide structure. The pressure plate is adaptively positioned and aligned with the test motherboard through the elastic guide structure. The drive unit is used to drive the pressure plate to move and press against the GPU under test, so that the GPU under test is pressed against the test motherboard. The pressing detection signal includes at least the pressing stroke of the pressure plate and / or the pressure value applied by the pressure plate to the GPU under test.

[0008] By adopting the above technical solution, the pressing drive structure was further optimized. The mechanical design employing a pressure plate and an elastic guide structure enables the pressure plate to have adaptive positioning and alignment capabilities, compensating for minor pose deviations between the GPU and the test motherboard, ensuring uniform pressure application. Simultaneously, the pressing detection signal is clearly defined to include at least the stroke and / or pressure value, providing direct and crucial feedback variables for closed-loop control, thus improving the consistency and reliability of the pressing process.

[0009] Preferably, the system also includes a heat sink, a circulating liquid cooling structure, and a heat dissipation detection module. The heat sink has a shaped cavity for supplying coolant flow. The heat sink is attached to the GPU under test to facilitate heat exchange with the GPU using coolant. The circulating liquid cooling structure is connected to the shaped cavity to circulate coolant into the cavity. Both the heat dissipation detection module and the circulating liquid cooling structure are electrically connected to the execution control module. The heat dissipation detection module is used to detect the heat dissipation operation status and / or leakage status of the circulating liquid cooling structure, and outputs the corresponding heat dissipation detection signal to the execution control module. The execution control module is used to control the operation status of the circulating liquid cooling structure based on the heat dissipation detection signal, so as to realize the heat dissipation regulation of the GPU under test.

[0010] By adopting the above technical solution, an active cooling system is integrated into the basic pressing fixture. Forced cooling of the GPU under test is achieved through a heatsink and circulating liquid cooling structure, ensuring the GPU's temperature stability under high load testing and preventing overheating-induced frequency throttling or damage that could affect the validity and integrity of the test results. The introduction of a thermal detection module further enables monitoring and closed-loop control of the cooling system's operating status (such as temperature and leakage), improving the safety and controllability of the testing process.

[0011] Secondly, this application provides a GPU testing system, including a test motherboard and any of the GPU testing fixtures described in the first aspect. The test motherboard is used to establish an electrical connection with the GPU under test, perform performance testing on the GPU under test and collect corresponding performance data, and the test motherboard establishes bidirectional communication with the execution control module of the GPU testing fixture through a preset external communication interface. The execution control module is used to acquire the performance data collected by the test motherboard and, in conjunction with the pressing detection signal from the pressing detection module, adaptively adjust the operating parameters of the pressing drive structure.

[0012] By adopting the above technical solution and establishing bidirectional communication between the test motherboard and the test fixture execution control module, the test fixture can obtain electrical performance test results from the test motherboard. The execution control module, combining the physical bonding state and electrical performance data, can adaptively adjust the bonding parameters. This means the system can reverse-engineer the front-end bonding process based on the final electrical connection quality, elevating quality control from the physical parameter level to the electrical performance level.

[0013] Preferably, the performance data collected by the test motherboard includes at least performance parameters used to evaluate the electrical connection quality between the GPU under test and the test motherboard; the execution control module has a pre-set anomaly diagnosis rule base, which defines the mapping relationship between different types of performance parameter anomalies and potential compression problems; The step of the execution control module adaptively adjusting the operating parameters of the pressing drive structure in conjunction with the pressing detection signal of the pressing detection module specifically includes: The execution control module is used to diagnose the performance data based on the anomaly diagnosis rule base and the current pressing detection signal, and determine whether there is a performance anomaly caused by the pressing state; if so, it determines the target pressing adjustment strategy based on the mapping relationship, and generates control commands to adjust the operating parameters of the pressing drive structure according to the target pressing adjustment strategy.

[0014] By adopting the above technical solution, a bonding adaptive adjustment method based on electrical performance is concretized. Through a pre-built anomaly diagnosis rule base, the system can correlate specific electrical performance anomaly patterns (such as high bit error rate, initialization failure) with potential bonding problems (such as insufficient local pressure, pin connection issues). When a performance anomaly caused by bonding is detected, the system can automatically trigger targeted bonding adjustment strategies (such as local pressure compensation, re-bonding), achieving an intelligent upgrade from fault detection to fault self-repair, significantly improving testing efficiency and first-test yield.

[0015] Preferably, the execution control module is used to obtain test plan information before the test motherboard performs performance testing on the GPU under test. The test plan information includes at least several test items that are executed sequentially during the performance test. The execution control module is also used to determine the expected heat dissipation requirements for the current test item to be executed and at least one subsequent test item in the test plan information in real time, and then, based on the expected heat dissipation requirements, to adaptively adjust the operating parameters of the circulating liquid cooling structure before the current test item to be executed begins.

[0016] By adopting the above technical solution, proactive heat dissipation control based on the test plan is achieved. The execution control module can predict the heat dissipation requirements of subsequent test loads by acquiring the test item sequence, and preemptively increase the heat dissipation capacity before the start of high-load test items. This feedforward control method avoids the response delay problem of traditional temperature control heat dissipation, effectively suppresses the temperature fluctuation of the GPU under test during the test, and provides a more stable and consistent thermal environment for performance testing, thereby ensuring the accuracy and comparability of test data.

[0017] Preferably, the heat dissipation operation status includes the contact surface temperature between the heat sink and the GPU under test; The execution control module is also used to receive the internal temperature data of the GPU under test fed back by the test motherboard in real time during the execution of any of the test items, and simultaneously acquire the contact surface temperature detected by the heat dissipation detection module; based on the internal temperature data and the contact surface temperature, estimate the actual thermal resistance state of the heat sink in real time, and when the actual thermal resistance state meets the warning conditions, generate and output a thermal resistance abnormality signal for the test personnel to know.

[0018] By adopting the above technical solution, a real-time method for diagnosing the health status of a heat dissipation system is provided. By comparing the core temperature inside the GPU under test with the contact surface temperature of the heat sink, the thermal resistance of the heat dissipation path can be calculated and monitored in real time. When the thermal resistance increases abnormally (possibly caused by aging of the thermal conductive material, poor contact, or blockage of the cold head), the system can issue a timely warning. This allows testers or the system to perform maintenance before a substantial decrease in heat dissipation performance leads to test failure or damage to the GPU under test, thus improving the reliability and maintainability of the system.

[0019] Preferably, the execution control module has several heat dissipation simulation configurations pre-stored, and each heat dissipation simulation configuration corresponds to a parameterized model of the heat dissipation performance of an external heat sink. The execution control module is used to, in response to the test motherboard's execution of a specific test item that requires simulating the heat dissipation performance of a target external heatsink, call the corresponding heat dissipation simulation configuration when the test motherboard executes the specific test item; determine the expected heat dissipation requirements for the execution process of the specific test item based on the heat dissipation performance parameterization model in the called heat dissipation simulation configuration; and adaptively adjust the operating state of the circulating liquid cooling structure according to the expected heat dissipation requirements so that the heat dissipation performance of the heat sink on the GPU under test simulates the target external heatsink, wherein the target external heatsink refers to a specific external heatsink.

[0020] By adopting the above technical solution, the functionality of the testing system is expanded, making it a heatsink performance simulation and verification platform. By pre-storing parametric models of different external heatsinks and calling the corresponding models to control the liquid cooling system according to the test items, this system can accurately simulate the heat dissipation performance of a GPU equipped with various target heatsinks in a laboratory environment. This allows R&D personnel to evaluate and screen heat dissipation solutions in advance during the GPU prototype stage without having to implement all candidate heatsinks, significantly shortening the product development cycle and reducing verification costs.

[0021] Preferably, the thermal performance parameterization model includes dynamic response parameters for simulating thermal inertia effects; the execution control module is used to control the operating state of the circulating liquid cooling structure in response to the power consumption or temperature changes of the tested GPU fed back by the test motherboard, according to the control logic defined by the dynamic response parameters, when executing the specific test item based on the thermal simulation configuration including the dynamic response parameters, so as to simulate the thermal inertia effect of the target external heat sink.

[0022] By adopting the above technical solutions, the realism and dimensionality of the thermal simulation are further enhanced. By introducing a model that includes dynamic response parameters, the system can simulate the thermal inertia effect of a real heatsink. In the test, when the load on the tested GPU changes abruptly, the response of the thermal system will exhibit a physically consistent delay, thus allowing for a more realistic assessment of the temperature and performance of the tested GPU under dynamic loads (such as computational bursts), providing more practically meaningful transient data for thermal design.

[0023] Preferably, the execution control module is further configured to, after executing a test based on a thermal simulation configuration and completing the test, acquire the actual temperature trajectory of the GPU under test recorded by the test motherboard during the entire test process, compare and analyze the actual temperature trajectory with the ideal temperature trajectory expected to be generated by the completed thermal simulation configuration, and automatically correct the parameters of the thermal simulation configuration based on the differences in the comparison analysis to generate an optimized thermal simulation configuration for subsequent testing.

[0024] By adopting the above technical solution, the system is endowed with the ability to self-learn and optimize the simulation model. By comparing the expected results of the heat dissipation simulation with the actual test temperature data, the system can automatically analyze the deviation and correct the simulation configuration parameters. This process enables the simulation model to continuously approximate the performance of the real heat sink, improving simulation accuracy. At the same time, it also reduces the dependence on the accuracy of the initial model, giving the system adaptive optimization capabilities for long-term use, thus enhancing the tool's practicality and value.

[0025] In summary, this application includes at least one of the following beneficial technical effects: The GPU testing fixture described in this application systematically solves the problems of unknown pressing state, fixed and rigid parameters, and insufficient heat dissipation under high load that exist in traditional testing fixtures through the collaborative design of closed-loop control architecture, adaptive mechanical structure and integrated thermal management, and achieves comprehensive control of GPU testing process with high precision, high reliability and high adaptability.

[0026] Furthermore, by establishing bidirectional communication between the test motherboard and the test fixture's execution control module, the test fixture can acquire electrical performance test results from the test motherboard. The execution control module, combining the physical bonding state with electrical performance data, can adaptively adjust the bonding parameters. This means the system can reverse-engineer the front-end bonding process based on the final electrical connection quality, elevating quality control from the physical parameter level to the electrical performance level. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the structure of a GPU testing fixture disclosed in Embodiment 1 of this application.

[0028] Figure 2 yes Figure 1 An exploded view of the GPU testing fixture.

[0029] Figure 3 This is a structural block diagram of a GPU testing fixture disclosed in Embodiment 1 of this application.

[0030] Figure 4 This is a structural block diagram of a GPU testing system disclosed in Embodiment 2 of this application.

[0031] Explanation of reference numerals in the attached drawings: 1. Pressing drive structure; 11. Drive unit; 111. Fixing frame; 112. Electric actuator; 113. Linear rail; 114. Movable plate; 115. Dust cover; 12. Pressure plate; 13. Elastic guide structure; 2. Pressing detection module; 21. Displacement sensor; 22. Pressure sensor; 3. Execution control module; 3. Heat sink; 31. Irregular cavity; 32. Cover plate; 33. Sealing ring; 4. Circulating liquid cooling structure; 41. Liquid supply tank; 42. Water pump; 43. Waterproof connector; 44. Liquid inlet pipe; 45. Liquid outlet pipe; 46. Water-cooled heat sink; 47. Cooling fan; 5. Heat dissipation detection module; 51. Temperature sensor; 52. Leakage sensor; 6. Test motherboard; 7. Host computer. Detailed Implementation

[0032] The following is in conjunction with the appendix Figure 1-4 This application will be described in further detail. Example

[0033] Embodiment 1 of this application discloses a GPU testing fixture. (Refer to...) Figure 1 , Figure 2 and Figure 3The GPU testing fixture includes a pressing drive structure 1, a pressing detection module 2, and an execution control module 3. The pressing drive structure 1 includes a drive unit 11, a pressure plate 12, and an elastic guide structure 13. The drive unit 11 specifically includes a fixed frame 111, an electric actuator 112 mounted on the fixed frame 111, a linear guide 113, and a movable plate 114 slidably connected to the fixed frame 111 along the length of the linear guide 113. A dust cover 115 is provided on the surface of the fixed frame 111. The drive end of the electric actuator 112 is connected to the movable plate 114 to drive the movable plate 114 to reciprocate along the length of the linear guide 113. The pressure plate 12 is made of a flexible material and is located on the sliding path of the movable plate 114. It contacts the movable plate 114 during its sliding process and is pressed down by the movable plate 114. A test main board 6 is pre-installed below the pressure plate 12. The pressure plate 12 and the test main board 6 are connected by an elastic guide structure 13. Specifically, the elastic guide structure 13 is a columnar structure with elastic deformation capability. The pressure plate 12 is fitted onto the elastic guide structure 13, and the lower end of the elastic guide structure 13 is inserted into a pre-installed slot in the test main board 6 to achieve the test main board... The positioning connection between the test motherboard 6 and the pressure plate 12 provides space for the GPU under test 8 to be inserted. When the pressure plate 12 is pressed down by the movable plate 114 and moves downward, the pressure plate 12 presses against the GPU under test during the downward movement, thereby pressing the GPU under test 8 into stable contact with the electrical interface (e.g., the PIN pin array or test socket in the card slot) of the test motherboard 6, thus establishing an electrical connection for power supply, signal transmission and performance testing. This electrical connection is the physical basis for the test motherboard 6 to apply test stimuli to the GPU under test 8 and collect its performance data.

[0034] Reference Figure 1 , Figure 2 and Figure 3The pressing detection module 2 includes a displacement sensor 21 for detecting the sliding stroke of the movable plate 114 along the length direction of the rail 113, and a pressure sensor 22 disposed on the lower surface of the movable plate 114. The pressure sensor 22 is used to detect the pressure value of the movable plate 114 pressing against the pressure plate 12 and the pressure plate 12 pressing down on the GPU 8 under test. The execution control module 3 is electrically connected to the electric push rod 112 and the pressing detection module 2 to receive the pressing detection signal detected by the pressing detection module 2. Accordingly, the pressing detection signal specifically includes the sliding stroke of the movable plate 114 (i.e., the pressing stroke of the pressure plate 12 and the GPU 8 under test), and the pressure value applied by the pressure plate 12 to the GPU 8 under test. In this embodiment, the execution control module 3 can be pre-set with an association lookup table. The association lookup table stores the standard pressing parameters corresponding to different GPU models. The standard pressing parameters include the corresponding standard displacement value and standard pressure value. Therefore, the execution control module 3 can find the standard pressing parameters corresponding to the current GPU 8 in the association lookup table according to the model of the current GPU 8 under test (which can be entered by the tester through a touch screen that has been pre-connected to the execution control module 3, so that the execution control module 3 knows the model of the current GPU 8 under test). The execution control module 3 determines in real time whether the pressing detection signal obtained in real time is consistent with the standard pressing parameters. When they are consistent, it means that the pressing state of the current pressing drive structure 1 has reached a stable pressing state. At this time, the GPU 8 under test and the test motherboard 6 have reached a connection state that can achieve stable testing. Then, the execution control module 3 controls the pressing drive structure 1 to stop moving, that is, to maintain the current state until the test ends (if the duration of maintaining the current state reaches a specified duration, the execution control module 3 controls the pressing drive structure 1 to reset and release the pressure on the GPU 8 under test).

[0035] Reference Figure 1 and Figure 2 Furthermore, the GPU testing fixture also includes a heat sink 3, a circulating liquid cooling structure 4, and a heat dissipation detection module 5, to dissipate heat from the GPU 8 under test and prevent excessive temperature rise during testing, which would affect the testing results. Specifically, the heat sink 3 is located between the movable plate 114 and the pressure plate 12, and the heat sink 3 is made of a material with good thermal conductivity (such as copper). The heat sink 3 has pre-milled irregular cavities 31 (such as S-shaped cavities) inside to store coolant and allow coolant flow.

[0036] Reference Figure 1 and Figure 2The circulating liquid cooling structure 4 includes a liquid supply tank 41, a water pump 42, a waterproof connector 43, an inlet pipe 44, an outlet pipe 45, a water-cooled radiator 46, and a cooling fan 47. The liquid supply tank 41 stores coolant. A through hole is pre-drilled at the bottom of the tank, connecting to the inlet of the water pump 42. The outlet of the water pump 42 connects to the inlet pipe 44, and the other end of the inlet pipe 44 connects to the inlet of the waterproof connector 43. Correspondingly, the irregular cavity 31 has two openings at its end. One opening connects to the inlet of the waterproof connector 43, and the other opening connects to the outlet of the waterproof connector 43. The other end of the outlet of the waterproof connector 43 connects to the outlet pipe 45, which connects to the water-cooled radiator 46. The inlet and outlet of the heat exchanger 46 are connected to the bottom of the liquid supply tank 41 via pipes. This heat exchanger 46 uses an existing structure, including liquid channels and cooling fins (i.e., thin metal sheets). The cooling fins quickly transfer heat from the hot water in the internal pipes to the air. The liquid channels connect to the inlet and outlet of the heat exchanger 46 to allow coolant to flow through. As the coolant flows through the liquid channels, it transfers heat to the cooling fins, achieving cooling before flowing out from the outlet. Screw holes are pre-drilled around the heat exchanger 46, and a cooling fan 47 is fixedly connected to it. The cooling fan 47 blows air towards the cooling fins to accelerate airflow and remove heat, thus cooling the coolant. Accordingly, the specific flow path of the coolant is as follows: the coolant is drawn from the supply tank 41 by the water pump 42, flows through the inlet pipe 44 and the waterproof connector 43 and then enters the irregular cavity 31, then flows out of the irregular cavity 31 to the outlet pipe 45, and enters the water-cooled heat sink 46 to dissipate heat before flowing back to the supply tank 41, thereby realizing the circulation and cooling flow of the coolant.

[0037] Reference Figure 1 , Figure 2 and Figure 3It should be noted that a cover plate 32 is covered on the surface of the heat sink 3, and a sealing ring 33 (such as fluororubber with good heat resistance is used as the sealing ring material) is provided on the surface of the heat sink 3 near the irregular cavity 31 and along the circumference of the irregular cavity 31 to improve the sealing performance of the joint between the cover plate 32 and the heat sink 3. The heat dissipation detection module 5 specifically includes a temperature sensor 51 and a leakage sensor 52. The temperature sensor 51 is located on the side wall of the cover plate 32 near the irregular cavity 31 to collect temperature data near the irregular cavity 31. The leakage sensor 52 is located below the waterproof connector 43 and between the waterproof connector 43 and the pressure plate 12 to detect whether there is coolant leakage at the waterproof connector 43, such as at the connection points between the waterproof connector 43 and the inlet pipe 44, the outlet pipe 45, or the opening of the irregular cavity 31. The leakage sensor 52 uses a single-layer copper-clad laminate with a loop-shaped coil milled on it. Once a leakage occurs, the leaking coolant will flow to the surface of the leakage sensor 52, which will be detected and a leakage signal will be emitted. Accordingly, the heat dissipation detection signal includes temperature data and leakage signal.

[0038] Reference Figure 1 , Figure 2 and Figure 3 The heat dissipation detection module 5, water pump 42, and cooling fan 47 are all electrically connected to the execution control module 3. The execution control module 3 is used to control the start and stop of water pump 42 and cooling fan 47, as well as the operating power of cooling fan 47, based on the temperature data detected by temperature sensor 51, thereby controlling the operating state of the circulating liquid cooling structure 4. Accordingly, in this embodiment, the execution control module 3 has preset operating states (i.e., on or off states) and operating parameters (such as water pump 42 speed and cooling fan 47 operating power) corresponding to different temperature ranges. By comparing the temperature data detected by temperature sensor 51 in real time with the temperature range, the current water pump 42 and cooling fan 47 are adjusted to a state consistent with the operating state and operating parameters of water pump 42 and cooling fan 47 corresponding to the current temperature range of temperature sensor 51. In addition, the execution control module 3 is also used to receive the leakage signal emitted by leakage sensor 52 and control the preset alarm (such as buzzer) to start the alarm.

[0039] The implementation principle of the GPU testing fixture disclosed in Embodiment 1 of this application is as follows: The GPU 8 under test is placed between the pressure plate 12 and the test motherboard 6. The monitoring personnel input the model of the GPU 8 under test using a preset touch screen, and start the execution control module 3 using a preset start button on the touch screen. The execution control module 3 obtains the corresponding standard pressing parameters based on the signal of the GPU 8 under test, and controls the drive unit 11 to start the downward movement of the movable plate 114, so that the movable plate 114 presses down the cover plate 32, the heat sink 3, and the pressure plate 12. During the downward movement, the pressure plate 12 presses the GPU 8 under test and attaches it to the surface of the test motherboard 6. During this period, the pressing detection module 2 is in direct contact with the cover plate 32 to detect the reverse pressure fed back to the movable plate 114 by the cover plate 32 after being pressed, thereby using this as the pressure plate 12. The pressure value applied by the pressure plate 12 to the GPU 8 under test is used as the displacement detection value of the displacement sensor 111 as the pressing stroke of the pressure plate 12 to the GPU 8 under test, thereby determining whether it is consistent with the standard pressing parameters. When consistent, it is considered that the GPU 8 under test and the test motherboard 6 have reached a connection state that can achieve stable testing. Then, the downward pressure of the driving unit 11 on the pressure plate 12 is paused and maintained for a specified time. In this application, the specified time is assumed to be the test time of the test motherboard 6 on the GPU 8 under test (this test time is pre-stored in the execution control module 3, and the execution control module 3 can pre-store the test time corresponding to different models of GPUs). When the pause time reaches the aforementioned specified time, the execution control module 3 controls the driving unit 11 to move upward, releasing the pressure of the pressure plate 12 on the GPU 8 under test, and completing the test. During the test, the temperature sensor 51 will detect the temperature data near the irregular cavity 31 in real time, and the execution control module 3 will adjust the operation status of the circulating liquid cooling structure 4 according to the real-time temperature data, so as to adjust the temperature of the coolant in the irregular cavity 31, thereby ensuring the heat dissipation effect of the coolant on the tested GPU 8. Example

[0040] This application further discloses a GPU testing system, referring to... Figure 3 and 4 The GPU testing system includes the GPU testing fixture described in Example 1, and also includes a test motherboard 6 for performing performance testing on the GPU under test 8. Specifically, the test motherboard 6 can be a dedicated test card that conforms to the PCIe interface specification or the specific GPU manufacturer's test specification. It integrates an electrical interface (such as a test socket, flexible probe array, or high-density connector) that matches the GPU pin layout. The test motherboard 6 makes stable contact with the pins of the GPU under test through the electrical interface, thereby achieving electrical connection with the GPU under test 8, and thus supplying power to the GPU, applying test stimulus signals, and collecting its response to complete the performance test of the GPU under test 8 and output the corresponding performance data.

[0041] Reference Figure 3 and 4 Specifically, both the test motherboard 6 and the execution control module 3 of the GPU test fixture are connected to a host computer 7 via a preset external communication interface (such as Ethernet, USB, or RS-485 bus). The host computer 7 acts as the central hub for data exchange and instruction scheduling, enabling indirect bidirectional communication and collaboration between the test motherboard 6 and the execution control module 3. The host computer 7 can be an industrial control computer or a high-performance industrial PC. Correspondingly, test personnel can perform operations such as entering the model of the GPU under test 8, starting the test process, and monitoring the test status through the human-machine interface of the host computer 7.

[0042] In addition, the host computer 7 has a large pre-stored test case library. This library defines corresponding test plan information for different GPU models, including a sequence of test items and detailed test content for each item. Test items include, but are not limited to: Basic test items include: power-on timing check, GPU device identification (Device ID), memory initialization and basic read / write test.

[0043] Functional test items: such as functional verification of each graphics computing unit (SM), video codec engine testing, and performance testing of specific AI operators.

[0044] Performance and stress test items: such as 3D graphics rendering benchmark tests (such as 3DMark), general computing tests (such as CUDA operations), and long-term high-load stress tests (such as FurMark).

[0045] After the tester selects the GPU8 signal under test on the host computer 7 interface, the host computer 7 first sends the GPU signal to the execution control module 3. As described in Example 1, the execution control module 3 determines the corresponding standard pressing parameters and starts the pressing drive structure 1. Combined with the feedback from the pressure detection module, the pressing state is made to reach a state consistent with the standard pressing parameters, thus completing the initial pressing closed-loop control.

[0046] After the initial compression is completed, the execution control module 3 sends a compression ready signal to the host computer 7. The host computer 7 then schedules the test motherboard 6 to execute the test items in sequence according to the test plan information corresponding to the GPU 8 under test. The test subject uploads the real-time collected performance data to the host computer 7 in real time, and the host computer 7 forwards it to the execution control module 3. Accordingly, the performance data collected and uploaded to the host computer 7 after the test motherboard 6 executes the above test items includes: the pass / fail status of each test item, the GPU operating frequency (core frequency, memory frequency), real-time power consumption, the reading of the internal temperature sensor 51 (such as the core temperature Tj), the completion time and accuracy of the computing task, the number of frames and errors of graphics rendering, and the bit error rate (BER) of the high-speed signal link, etc.

[0047] The execution control module 3 has a pre-built anomaly diagnosis rule base, which defines the mapping relationship between specific types of performance data anomalies and suggested compression adjustment directions or strategies; this mapping relationship points to the overall compression parameter adjustment, for example: Mapping Relationship A: If the bit error rate (BER) of a specific high-speed signal channel (such as PCIe Lane) continues to exceed the preset threshold Th_BER (e.g., 1E-12) for T1 seconds, it is mapped as follows: the overall pressing pressure may be insufficient or the contact impedance is too high, resulting in a decrease in signal quality; the recommended strategy is to increase the target holding pressure by a preset increment ΔF1 (e.g., 10N) at the current position.

[0048] Mapping Relationship B: If the test motherboard 6 receives a power-on failure or device unresponsive hardware status signal N times (e.g., 3 times) consecutively when attempting to power on and initialize the GPU core, it is mapped as follows: there is poor contact on the critical power pin or control pin, resulting in power supply or communication interruption; the recommended strategy is to perform a complete release and re-press cycle (i.e., execute the control module 3 to control the pressing drive structure 1 to reset and then re-execute the pressing operation) to reshape the contact interface.

[0049] Mapping Relationship C: In continuous load testing, the standard deviation of the GPU instantaneous power consumption value collected by the test motherboard 6 exceeds the stability threshold Th_σ. At the same time, the standard deviation of the real-time pressure value F_now in the pressure detection signal also exceeds the associated threshold Th_σF. This is mapped to: there is micro-instability in the pressure contact, which is manifested as the electrical contact resistance fluctuating synchronously with the pressure. The recommended strategy is to fine-tune the target value of the pressure control closed loop from F_target to F_target+ΔF2 (e.g., 5N) to enhance the static contact force and suppress micro-vibrations.

[0050] The execution control module 3 combines the current real-time compression detection signal (i.e., the current pressure F_now and stroke S_now) with the mapping relationship in the above-mentioned anomaly diagnosis rule base to make a comprehensive judgment. For example, even if the BER exceeds the standard, if the current F_now is already high and stable, the cause of "insufficient pressure" may be ruled out, and other problems (such as GPU defects or test motherboard 6 failure) may be considered, thereby avoiding incorrect adjustments. When the execution control module 3 determines that the performance abnormality is indeed caused by a compression problem, it directly generates a specific and executable target compression adjustment strategy based on the suggested strategy mapped from the rule base. This target compression adjustment strategy is converted into a direct control instruction for the compression drive structure 1. For example, if the suggested strategy is to increase the overall holding pressure ΔF, the target compression adjustment strategy is to adjust the pressure control target value from the standard pressure value F_target corresponding to the standard compression parameters to F_target+ΔF, and the drive unit 11 executes it. If the suggested strategy is to perform loosening and re-compression, the target compression adjustment strategy is to control the drive unit 11 to reset to the starting point and re-execute the compression action with the standard compression parameters as the target. After the adjustment is completed, the execution control module 3 notifies the host computer 7 to complete the adjustment.

[0051] Optionally, before starting the test process, the host computer 7 sends the determined test plan information to the execution control module 3, and the test plan information is specifically presented as a structured list, such as: [Test item 1: ID recognition, estimated duration: 10 seconds; Test item 2: Light load diagnosis, estimated duration: 30 seconds; Test item 3: Stress test (FurMark), estimated duration: 600 seconds].

[0052] The execution control module 3 has a pre-stored lookup table mapping test items to heat dissipation requirements. This table records the typical power consumption or heat generation levels of various test items. Upon receiving the test plan information, the execution control module 3 queries the corresponding heat dissipation requirement level for the currently pending test item (i.e., the first pending test item) and for each subsequent test item. For example: If the test item is "light load diagnostics," the corresponding heat dissipation requirement level is low. If the test item is "stress test," the requirement level is high. This is used to determine the heat dissipation requirement level (i.e., predicted heat dissipation requirement) for all test items in the test plan information.

[0053] Then, the execution control module 3, based on the determined heat dissipation requirement level, generates and executes a pre-adjustment instruction at a specific moment before the test item is executed (e.g., T_pre = 5 seconds in advance). It should be noted that the execution control module 3 pre-stores the operating parameters of the circulating liquid cooling structure 4 corresponding to different heat dissipation requirement levels (wherein, the operating state is specifically manifested as the speed of the water pump 42 and / or the working power of the cooling fan 47). Therefore, the execution control module 3 can generate the pre-adjustment instruction based on the above-mentioned pre-stored correspondence. An exemplary pre-adjustment instruction can be: 5 seconds before the pressure test item starts, the execution control module 3 sends an instruction to the circulating liquid cooling structure 4 to increase the cooling pump speed from a low level (e.g., 2000 RPM) to a high level (e.g., 5000 RPM), thereby achieving adaptive adjustment of the operating parameters of the circulating liquid cooling structure 4.

[0054] Furthermore, during the test execution, the test motherboard 6 periodically reads the reported core temperature value Tj from the internal sensor registers of the GPU under test 8 by calling debugging instructions or driver interfaces provided by the GPU manufacturer (such as NVML for NVIDIA GPU, ROCm-SMI for AMD GPU). This core temperature value Tj is then forwarded to the execution control module 3 via the host computer 7300 through the operating system or embedded software of the test motherboard 6.

[0055] The execution control module 3 obtains the GPU core temperature value Tj read by the test motherboard 6 through the host computer 7, and simultaneously reads the contact surface temperature Tc of the heat sink 3 measured by the heat dissipation detection module 5 (it should be noted here that the temperature data near the irregular cavity 31 described in the default embodiment 1 is equivalent to the contact surface temperature between the heat sink 3 and the GPU 8 under test. Therefore, the contact surface temperature of the heat sink 3 here is the temperature data detected by the temperature sensor 51 included in the heat dissipation detection module 5); at the same time, it obtains the real-time power consumption P of the GPU from the data of the test motherboard 6 or internal estimation. Specifically, the real-time power consumption P of the GPU is obtained as follows: the test motherboard 6 synchronously reads the chip real-time power consumption estimate reported by the internal power management unit (PMU) of the GPU through the above-mentioned debugging interface, and uses it as the real-time power consumption P of the GPU. Alternatively, a high-precision current detection circuit (such as a sampling resistor + differential amplifier) ​​and a voltage sampling circuit are set on the GPU power supply circuit (such as the input path of Vcore and Vmem) of the test motherboard 6. The controller (such as FPGA or MCU) of the test motherboard 6 collects these data at a high sampling rate and calculates the instantaneous power consumption P_calculated=Σ(V_bus_n*I_bus_n) in real time. Then, the calculation result is sent to the execution control module 3 as the real-time power consumption P of the GPU.

[0056] The execution control module 3 aligns the GPU core temperature Tj, the heatsink 3 contact surface temperature Tc, and the GPU real-time power consumption P in time. Then, when the GPU is under steady-state test load (e.g., power consumption P fluctuates less than ±5% over N consecutive sampling periods), the execution control module 3 calculates the approximate thermal resistance under this steady-state condition: Rth_calc = (Tj - Tc) / P, and updates this instantaneous approximate thermal resistance value in real time. The instantaneous thermal resistance approximate value Rth_calc is used to characterize the actual thermal resistance state of the heatsink 3. The execution control module 3 has pre-set warning conditions, such as: Condition 1: Under steady-state test load, Rth_calc exceeds the preset threshold Rth_max (e.g., 0.5 °C / W) for more than the preset T_warn seconds (e.g., 30 seconds). Condition 2: Over multiple consecutive steady-state test load periods, the calculated Rth_calc, compared to the initial test value or a preset historical benchmark value of the same model, suddenly increases by more than a proportion Δx (e.g., 30%) (e.g., within a specified time). When any of the aforementioned conditions are met, the execution control module 3 determines that the thermal resistance is abnormal, generates an early warning signal, and displays the early warning signal through the human-machine interface of the host computer 7. The early warning signal may be: Alarm: The thermal resistance of the heat dissipation system has increased abnormally. Possible causes: aging of the thermal conductive material, poor contact, or blockage of the waterproof connector 43.

[0057] Optionally, during the process of the host computer 7 scheduling the test motherboard 6 to execute test items sequentially, if there is a test item named "Cooling Simulation Test" (hereinafter referred to as a specific test item) in the test plan information, when it is the turn to execute the specific test item, the host computer 7 sends a simulation command to the execution control module 3 before scheduling the test motherboard 6 to execute the specific test item. For example: start simulation mode, configuration: heatsink A (hereinafter referred to as the target external heatsink), test load: steady state 250W (FurMark 4K). It should be noted here that the specific test item includes the specific external heatsink that needs to be configured, such as heatsink A in this case.

[0058] After receiving the aforementioned simulation command, the execution control module 3 retrieves the thermal performance parameterized model corresponding to the target external heat sink from its internally pre-stored thermal simulation configuration library. The thermal simulation configuration library stores thermal simulation configurations for different external heat sinks, which are specifically represented by thermal performance parameterized models. These models are represented by a set of parameters, for example: {Maximum heat dissipation power: 250W, Thermal resistance curve: Rth=0.3+0.1 / Flow (where Flow is the equivalent airflow)}.

[0059] The execution control module 3 calculates the equivalent thermal resistance target value (Rth_target) to be simulated based on the thermal performance parameterized model corresponding to the current target external heatsink and the current test load (here, the current test load refers to the standardized load that the test motherboard 6 plans to apply to the GPU under test 8 during the upcoming thermal simulation test, used to evaluate the thermal performance; this load information is issued by the host computer 7 as part of the metadata of a specific test item; the test load given in the aforementioned simulation instruction means that in the simulation test, the test motherboard 6 will run the FurMark 4K test to stabilize the GPU power consumption at approximately 250W). The specific calculation method is as follows: the execution control module 3 queries the thermal resistance curve Rth=0.3+0.1 / Flow in the target external heatsink based on the obtained load information (e.g., P_load=250W). Here, a key setting is required: the simulation must assume that the target heatsink is working under its typical or rated conditions. For example, it is assumed that the equivalent airflow Flow_rated=1.0 (dimensionless unit) of the target external heatsink at its maximum speed. Then, substitute Flow=1.0 and P_load=250W into the thermal performance parameterization model: First, calculate the thermal resistance Rth_target = 0.3 + 0.1 / 1.0 = 0.4 °C / W. Rth_target represents the thermal resistance performance required of the target external heatsink under a 250W load to control the temperature rise of the tested GPU8 to the theoretical value (relative to ambient temperature) ΔT = P_load * Rth_target = 250W * 0.4 °C / W = 100°C.

[0060] Then, the execution control module 3 sends a signal to the host computer 7 indicating that the simulation mode is in place. Next, after the test motherboard 6 completes the test item in the test item sequence that precedes the specific test item, it enters the collaborative testing phase. The host computer 7 synchronously schedules the test motherboard 6 to execute the specific test item and schedules the execution control module 3 to execute closed-loop simulation control. The closed-loop simulation control process is as follows: During the execution of a specific test item, the execution control module 3 executes the aforementioned thermal resistance calculation process at a fixed cycle (e.g., per second): it obtains the real-time power consumption P_now and GPU core temperature Tj_now reported by the test motherboard 6 through the host computer 7; it reads the contact surface temperature Tc_now of the heat sink 3 measured by the heat dissipation detection module 5, and calculates the transient thermal resistance: Rth_inst=(Tj_now-Tc_now) / P_now.

[0061] Then, the proportional-integral control process is executed by control module 3: Rth_inst and Rth_target are compared, and the speed of the water pump 42 in the circulating liquid cooling structure 4 is dynamically adjusted based on the difference (e = Rth_inst - Rth_target): If e > 0 (the measured thermal resistance is too high, meaning the actual heat dissipation is weaker than the target radiator), the speed of the water pump 42 is increased, thereby increasing the coolant flow rate to enhance heat dissipation, thus causing Rth_inst to decrease; if e < 0 (the measured thermal resistance is too low, meaning the actual heat dissipation is stronger than the target radiator), the speed of the water pump 42 is decreased, thereby decreasing the coolant flow rate to weaken heat dissipation, thus causing Rth_inst to increase. The magnitude of the adjustment is proportional to the error e.

[0062] Furthermore, the thermal performance parameterization model also includes dynamic response parameters, which define at least one key indicator: the thermal time constant (τ, unit: seconds). For example, the thermal performance parameterization model is extended to: {maximum heat dissipation power: 250W, steady-state thermal resistance curve: Rth_ss = 0.3 + 0.1 / Flow, thermal time constant: τ = 15 seconds}. The larger the τ value, the greater the thermal inertia of the external heatsink and the slower the response. During the execution of the simulation test item based on this thermal performance parameterization model, the execution control module 3 adds the following dynamic processing step to the closed-loop simulation control flow described above: the execution control module 3 continuously monitors the real-time power consumption (P_now) of the GPU obtained from the test motherboard 6. When it is detected that the absolute value of the change in P_now within two consecutive sampling periods, |ΔP|, exceeds the preset step threshold (e.g., ΔP_threshold = 50W), and then stabilizes at the new power consumption platform (fluctuation < 5%) for at least several seconds, it is determined that a power consumption step change has occurred. Record the steady-state power consumption P_old before the change and the new steady-state power consumption P_new after the change.

[0063] Based on the transient thermal resistance calculation formula described above, calculate the steady-state target thermal resistances Rth_ss_old and Rth_ss_new corresponding to the steady-state power consumption P_old before the change and the new steady-state power consumption P_new after the change. For example: P_old = 100W → Rth_ss_old = 0.4 °C / W; P_new = 250W → Rth_ss_new = 0.4 °C / W.

[0064] Starting from the moment of the step change (t=0), the execution control module 3 no longer directly sets the tracking target of the closed-loop control to Rth_ss_new, but instead generates a dynamic target value Rth_target(t) that changes over time. Its variation follows the pattern of a first-order inertial element: Rth_target(t) = Rth_ss_new + (Rth_ss_old - Rth_ss_new) * exp(-t / τ); where τ is the thermal time constant (15 seconds) in the parameterized model of heat dissipation performance. This means that Rth_target starts from Rth_ss_old and decays exponentially, approaching Rth_ss_new infinitely after about 3τ (45 seconds).

[0065] The closed-loop simulation process of the execution control module 3 remains unchanged (i.e., the proportional-integral control process described above), but the target value it tracks is replaced from the fixed Rth_target to the time-varying Rth_target(t) mentioned above. The execution control module 3 calculates the current Rth_target(t) every second, then compares it with the measured Rth_inst, generating an error e(t) = Rth_inst - Rth_target(t), and adjusts the coolant flow rate accordingly, and adjusts the speed of the water pump 42 accordingly, thereby adjusting the coolant flow rate.

[0066] Because the control target Rth_target(t) changes slowly, the adjustment of coolant flow by the execution control module 3 will also be gradual and lagging. This results in the cooling system not immediately cooling at maximum capacity when the GPU power consumption jumps from 100W to 250W. The GPU core temperature (Tj) will rise at a rate and with an overshoot that conforms to the physical characteristics of a real heatsink A (τ=15 seconds). By observing the GPU's frequency-temperature curve over time under this simulated dynamic cooling condition, its real-world performance when encountering a cooling bottleneck in actual use (such as game loading, bursts of computational tasks) can be evaluated.

[0067] Optionally, after completing specific test items for the target external heatsink, the execution control module 3 obtains the following two sets of time series data recorded by the test motherboard 6 through the host computer 7: GPU actual power trajectory (P_actual(t)): The real-time power consumption array of the GPU throughout the entire test, with a sampling interval of Δt.

[0068] Actual core temperature trajectory (Tj_actual(t)): An array of GPU core temperatures sampled in sync with the GPU’s real-time power consumption.

[0069] The execution control module 3 utilizes the thermal performance parameterization model (denoted as Model_old, e.g., {Rth_ss_old, τ_old, ...}) invoked in this round of testing, combined with the obtained P_actual(t), to re-simulate "what the GPU's internal core temperature should be if the target external heatsink operates strictly according to this thermal performance parameterization model". The process of calculating the GPU's internal core temperature is a forward solving process, the specific steps of which are as follows: For each time point t_i: Calculate the instantaneous steady-state thermal resistance Rth_ss(t_i) of the target external heatsink at that power consumption, based on the power consumption P_actual(t_i) and the steady-state thermal resistance curve in the model Model_old. If the model contains a dynamic parameter τ, Rth_ss(t_i) cannot be used directly. A dynamic target thermal resistance Rth_target(t_i) needs to be calculated based on the thermal inertia response defined in the model (e.g., first-order hysteresis), considering the power consumption history. This calculation logic must be completely consistent with the dynamic target thermal resistance generation steps described above, forming a closed loop. Next, assuming the heatsink performance exactly reaches Rth_target(t_i), the temperature difference between the GPU chip junction temperature and the heatsink base (i.e., the contact surface of heatsink 3) should be: ΔT_model(t_i) = P_actual(t_i) * Rth_target(t_i). To obtain an absolute junction temperature estimate, a reference temperature needs to be added. Here, it is assumed that the radiator base temperature (Tc) remains at a constant reference value during the simulation (e.g., Tc_ref = 40°C, which can be set). The formula for calculating the expected temperature trajectory is: Tj_model(t_i) = Tc_ref + ΔT_model(t_i) = Tc_ref + P_actual(t_i) * Rth_target(t_i). Performing this calculation over all time points yields the complete Tj_model(t) array.

[0070] The execution control module 3 compares the calculated Tj_model(t) array with the measured Tj_actual(t) array point by point, and calculates two key error indicators: 1. Steady-state error (ΔT_ss): Select a period during the test where the GPU power consumption is stable (e.g., the last 300 seconds), and calculate the average value of Tj_actual(t) minus the average value of Tj_model(t) during this period. ΔT_ss = mean(Tj_actual) - mean(Tj_model).

[0071] 2. Dynamic Fit Error (RMSE): Calculates the root mean square value of the difference between the two over the entire test period to evaluate the overall waveform fit. RMSE = sqrt(mean([Tj_actual(t) - Tj_model(t)]^2)).

[0072] Finally, based on the calculated error, the internal parameters of the model Model_old are automatically modified according to preset mathematical rules. For example, when |ΔT_ss|>2.0°C or RMSE>5.0°C, the parameterized model of heat dissipation performance is considered to have a significant deviation, triggering the correction process.

[0073] The proposed changes are as follows: Modification of the steady-state thermal resistance parameter Rth_ss: If ΔT_ss > 0 (the measured temperature is higher than the model prediction, i.e., the thermal performance parameterization model is too optimistic and the heat dissipation capacity is overestimated), then the thermal resistance of the thermal performance parameterization model needs to be increased. The modified formula is: Rth_ss_new = Rth_ss_old * (1 + α * ΔT_ss / mean(Tj_actual)), where α is a preset learning rate coefficient (e.g., α = 0.5) used to control the magnitude of a single adjustment and prevent over-adjustment.

[0074] Modification of the thermal time constant τ: If the dynamic error RMSE is large, and analysis of the error curve reveals a phase lag in the temperature rise / fall, τ can be adjusted. For example, if the measured temperature rises more slowly than the model predicts, τ needs to be increased. τ_new = τ_old * (1 + β * (RMSE_delay / t_scale)); where β is another learning rate, RMSE_delay is the root mean square value focused on phase error calculation, and t_scale is the time scale normalization factor.

[0075] After applying the modified formula, a new set of parameters {Rth_ss_new, τ_new, ...} is obtained. The execution control module 3 saves this set of parameters as a new heat dissipation simulation configuration and associates it with the target external heat sink. The next time the test plan information calls "external heat sink A" for simulation, the execution control module 3 will prioritize calling the new heat dissipation simulation configuration. Because the parameters of this new heat dissipation simulation configuration are modified based on the previous actual test data, its simulated temperature trajectory Tj_model(t) will be closer to the next actual measurement Tj_actual(t), thereby achieving iterative improvement in the accuracy of the parameterized model of heat dissipation performance.

[0076] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A GPU testing fixture, characterized in that: include: The pressure-driving structure (1) is used to apply a preset pressure to the GPU under test so that the GPU under test is precisely attached to the preset test motherboard (6); The pressing detection module (2) is used to detect the pressing state of the pressing drive structure (1) and output the corresponding pressing detection signal; The execution control module (3) is electrically connected to the pressing drive structure (1) and the pressing detection module (2) respectively, and is used to receive the pressing detection signal from the pressing detection module (2) and control the pressing action of the pressing drive structure (1) based on the pressing detection signal.

2. The GPU testing fixture according to claim 1, characterized in that: The pressing drive structure (1) includes a drive unit (11), a pressure plate (12), and an elastic guide structure (13). The pressure plate (12) is adaptively positioned and aligned with the test motherboard (6) through the elastic guide structure (13). The drive unit (11) is used to drive the pressure plate (12) to move and press against the GPU under test, so that the GPU under test is pressed against the test motherboard (6). The pressing detection signal includes at least the pressing stroke of the pressure plate (12) and / or the pressure value applied by the pressure plate (12) to the GPU under test.

3. The GPU testing fixture according to claim 1, characterized in that: It also includes a heat sink (3), a circulating liquid cooling structure (4) and a heat dissipation detection module (5). The heat sink (3) has a shaped cavity (31) for the flow of coolant. The heat sink (3) is attached to the GPU under test to achieve heat exchange with the GPU under test using coolant. The circulating liquid cooling structure (4) is connected to the irregular cavity (31) for circulating supply of coolant to the irregular cavity (31); The heat dissipation detection module (5) and the circulating liquid cooling structure (4) are both electrically connected to the execution control module (3). The heat dissipation detection module (5) is used to detect the heat dissipation operation status and / or leakage status of the circulating liquid cooling structure (4) and output the corresponding heat dissipation detection signal to the execution control module (3). The execution control module (3) is used to control the operation status of the circulating liquid cooling structure (4) based on the heat dissipation detection signal to realize the heat dissipation regulation of the GPU under test.

4. A GPU testing system, comprising a test motherboard (6) and a GPU testing fixture as described in any one of claims 1-3, characterized in that: The test motherboard (6) is used to establish an electrical connection with the GPU under test, perform performance tests on the GPU under test and collect corresponding performance data, and the test motherboard (6) establishes bidirectional communication with the execution control module (3) of the GPU test fixture through a preset external communication interface; The execution control module (3) is used to acquire the performance data collected by the test motherboard (6) and, in conjunction with the pressing detection signal of the pressing detection module (2), adaptively adjust the operating parameters of the pressing drive structure (1).

5. The GPU testing system according to claim 4, characterized in that: The performance data collected by the test motherboard (6) includes at least the performance parameters used to evaluate the electrical connection quality between the GPU under test and the test motherboard (6); the execution control module (3) has a pre-set abnormal diagnosis rule library, which defines the mapping relationship between different types of performance parameter abnormalities and potential compression problems; The specific steps of the execution control module (3) adaptively adjusting the operating parameters of the pressing drive structure (1) in conjunction with the pressing detection signal of the pressing detection module (2) include: The execution control module (3) is used to diagnose the performance data based on the abnormality diagnosis rule base and the current pressing detection signal, and to determine whether there is a performance abnormality caused by the pressing state; if so, the target pressing adjustment strategy is determined according to the mapping relationship, and control instructions are generated according to the target pressing adjustment strategy to adjust the operating parameters of the pressing drive structure (1).

6. The GPU testing system according to claim 4, characterized in that: The GPU testing fixture is the GPU testing fixture as described in claim 3; the execution control module (3) is used to obtain test plan information before the test motherboard (6) performs performance testing on the GPU under test, and the test plan information includes at least several test items to be executed sequentially during the performance test. The execution control module (3) is also used to determine the expected heat dissipation requirements for the current test item to be executed and at least one subsequent test item in the test plan information in real time, and then, based on the expected heat dissipation requirements, to adaptively adjust the operating parameters of the circulating liquid cooling structure (4) before the current test item to be executed begins.

7. The GPU testing system according to claim 6, characterized in that: The heat dissipation operation status includes the contact surface temperature between the heat sink (3) and the GPU under test; The execution control module (3) is also used to receive the internal temperature data of the GPU under test fed back by the test motherboard (6) in real time during the execution of any of the test items, and simultaneously obtain the contact surface temperature detected by the heat dissipation detection module (5); based on the internal temperature data and the contact surface temperature, estimate the actual thermal resistance state of the heat sink (3) in real time, and generate and output a thermal resistance abnormality signal when the actual thermal resistance state meets the warning conditions, so that the test personnel can be informed.

8. The GPU testing system according to claim 4, characterized in that: The GPU testing fixture is the GPU testing fixture as described in claim 3; The execution control module (3) has several heat dissipation simulation configurations pre-stored, and each heat dissipation simulation configuration corresponds to a parameterized model of the heat dissipation performance of an external heat sink. The execution control module (3) is used to, in response to the execution operation of the test motherboard (6) on the specific test item that requires simulating the heat dissipation performance of the target external heat sink, call the corresponding heat dissipation simulation configuration when the test motherboard (6) executes the specific test item; determine the expected heat dissipation requirements for the execution process of the specific test item based on the heat dissipation performance parameterization model in the called heat dissipation simulation configuration; and adaptively adjust the operating state of the circulating liquid cooling structure (4) according to the expected heat dissipation requirements so that the heat sink (3) simulates the heat dissipation performance of the GPU under test to the target external heat sink, wherein the target external heat sink refers to a certain external heat sink.

9. The GPU testing system according to claim 8, characterized in that: The heat dissipation performance parameterization model includes dynamic response parameters for simulating thermal inertia effects; the execution control module (3) is used to control the operating state of the circulating liquid cooling structure (4) in response to the power consumption or temperature change of the tested GPU fed back by the test motherboard (6) according to the control logic defined by the dynamic response parameters when executing the specific test item based on the heat dissipation simulation configuration containing the dynamic response parameters, so as to simulate the thermal inertia effect of the target external heat sink.

10. The GPU testing system according to claim 8 or 9, characterized in that: The execution control module (3) is also used to obtain the actual temperature trajectory of the GPU under test recorded by the test motherboard (6) during the entire test process after the test is performed based on a heat dissipation simulation configuration, compare and analyze the actual temperature trajectory with the ideal temperature trajectory expected to be generated by the completed heat dissipation simulation configuration, and automatically correct the parameters of the heat dissipation simulation configuration according to the difference in the comparison analysis to generate an optimized heat dissipation simulation configuration for subsequent testing.

Images