Graphic processor detection method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of servers, and discloses a graphics processor detection method and device, computer equipment and a storage medium, and the method comprises the steps: carrying out the full-pressure pressurization of a to-be-detected graphics processor in a to-be-detected server through a graphics processor pressurization tool in a server aging link of a factory end; performing performance testing on the to-be-tested graphics processor to obtain a test value of a performance index of the to-be-tested graphics processor; acquiring the working environment temperature of the to-be-tested graphics processor; screening out a reference value of a performance index of the graphics processor corresponding to the working environment temperature from a target mapping table, wherein the target mapping table comprises a corresponding relationship between different working environment temperatures and the reference value of the performance index of the graphics processor; determining a deviation ratio based on the test value and the reference value of the performance index; and under the condition that the deviation ratio exceeds the target deviation ratio threshold, determining that the to-be-detected graphics processor has a risk and the detection is not passed. The detection accuracy of the graphics processor is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of servers, and in particular to a graphics processor detection method, device, computer equipment and storage medium. Background Art

[0002] With the rapid development of high-performance computing, the importance of graphics processing units (GPUs) in data centers and various computing tasks has become increasingly prominent. However, the high performance of GPUs also brings about high temperature problems, resulting in performance degradation and the risk of hardware damage.

[0003] In order to reduce the high temperature problem caused by the high performance of the GPU, the relevant technology intercepts GPUs with potential over-temperature risks before the artificial intelligence (AI) server leaves the factory. Specifically, the relevant technology places the AI ​​server to be tested in a high-temperature aging chamber, sets the ambient temperature of the high-temperature aging chamber to 38 degrees Celsius, uses the GPU pressurization tool to fully pressurize the GPU of the AI ​​server to be tested, uses the baseboard management controller (BMC) to monitor the temperature of the GPU in real time, and triggers an alarm when the temperature of the GPU exceeds the threshold, so as to detect whether the GPU has a potential over-temperature risk.

[0004] This detection method relies on a single temperature threshold to determine whether the GPU has a potential over-temperature risk. It is very likely that the GPU temperature will briefly exceed the temperature threshold, but there is no real over-temperature risk, resulting in a low accuracy rate of the detection result. Summary of the invention

[0005] In view of this, the present invention provides a graphics processor detection method, apparatus, computer equipment and storage medium to solve the problem that the detection method in the related art relies on a single temperature threshold to determine whether a GPU has a potential over-temperature risk. It is very likely that the temperature of the GPU briefly exceeds the temperature threshold, but there is actually no real over-temperature risk, resulting in a low accuracy of the detection result.

[0006] In a first aspect, the present invention provides a graphics processor detection method, the method comprising:

[0007] In the server aging phase at the factory end, the GPU stressing tool is used to fully stress the GPU under test in the server under test;

[0008] Performing a performance test on the graphics processor to be tested to obtain a test value of a performance indicator of the graphics processor to be tested;

[0009] Obtaining the working environment temperature of the graphics processor to be tested;

[0010] Filtering out a reference value of a performance indicator of a graphics processor corresponding to the working environment temperature from a target mapping table, wherein the target mapping table includes a correspondence between different working environment temperatures and reference values ​​of performance indicators of a graphics processor;

[0011] Determining a deviation ratio based on the test value and the benchmark value of the performance indicator;

[0012] When the deviation ratio exceeds the target deviation ratio threshold, it is determined that the graphics processor to be tested has a risk and the test fails.

[0013] The graphics processor detection method provided in the embodiment uses a graphics processor pressurization tool to perform full pressure pressurization on the graphics processor to be tested in the server to be tested during the server aging stage at the factory end, performs a performance test on the graphics processor to be tested, obtains a test value of the performance index of the graphics processor to be tested, obtains the working environment temperature of the graphics processor to be tested, screens out a reference value of the performance index of the graphics processor corresponding to the working environment temperature from a target mapping table, determines a deviation ratio based on the test value and the reference value of the performance index, and then determines a detection result of the graphics processor to be tested based on a comparison between the deviation ratio and a target deviation ratio threshold, thereby improving the accuracy of the detection result and enabling earlier fault detection.

[0014] In an optional implementation, obtaining the working environment temperature of the graphics processor to be tested includes:

[0015] Based on the baseboard management controller, obtaining the air inlet sensor temperature of the server to be tested;

[0016] Based on the air inlet sensor temperature of the server to be tested, the operating environment temperature of the graphics processor to be tested is determined.

[0017] The graphics processor detection method provided in this embodiment obtains the air inlet sensor temperature of the server to be tested based on the baseboard management controller, and then determines the working environment temperature of the graphics processor to be tested, thereby realizing the detection of the graphics processor to be tested under room temperature environment at the factory end. The graphics processor can be detected without using a high-temperature aging chamber, saving costs and reducing power consumption.

[0018] In an optional implementation, before filtering out the reference value of the performance indicator of the graphics processor corresponding to the working environment temperature from the target mapping table, the method further includes:

[0019] Placing a plurality of normal servers in a constant temperature box, and setting the working environment temperature of the constant temperature box to a first temperature;

[0020] When the duration of the working environment temperature of the thermostat at the first temperature reaches a duration threshold, using a graphics processor pressurization tool to perform full pressure pressurization processing on graphics processors in multiple normal servers;

[0021] For any normal server, multiple performance tests are performed on the graphics processors in different slots of the normal server to obtain multiple performance test results of the graphics processors in different slots;

[0022] The performance test results of multiple GPUs in the same slot in multiple normal servers are averaged to obtain the benchmark values ​​of the performance indicators of GPUs in different slots.

[0023] Determine the next temperature to be measured based on the order of the temperature values ​​in the temperature range to be measured;

[0024] The next temperature to be measured is determined as the first temperature, and the step of setting the working environment temperature of the thermostat to the first temperature is returned to execute until all temperature values ​​in the temperature range to be measured are tested, and the benchmark values ​​of the performance indicators of the graphics processors in different slots under different working environment temperatures are obtained;

[0025] Determine a mapping table corresponding to the graphics processor in each slot based on the benchmark values ​​of the performance indicators of the graphics processors in different slots at different working environment temperatures;

[0026] The target mapping table is selected from the plurality of mapping tables based on the slot where the graphics processor to be tested is located.

[0027] The graphics processor detection method provided in this embodiment places multiple normal servers in a constant temperature box and sets different working environment temperatures, performs full pressure pressurization treatment and performance testing on the graphics processors of the multiple normal servers, obtains benchmark values ​​of performance indicators of graphics processors in different slots under different working environment temperatures, ensures the accuracy of the benchmark values ​​of the performance indicators obtained, and then detects whether the graphics processor to be tested has a potential over-temperature risk based on the obtained benchmark values ​​of the performance indicators, thereby improving the accuracy of the detection results of the graphics processor.

[0028] In an optional implementation, determining the deviation ratio based on the test value and the reference value of the performance indicator includes:

[0029] The deviation ratio is determined by the following formula:

[0030]

[0031] The graphics processor detection method provided in this embodiment determines the deviation ratio according to the test value and the reference value of the performance indicator, and then determines the detection result of the graphics processor according to the comparison between the deviation ratio and the target deviation ratio threshold, thereby improving the accuracy of the detection result.

[0032] In an optional embodiment, the method further includes:

[0033] When the deviation ratio exceeds a target deviation ratio threshold, an alarm is triggered and the graphics processor to be tested is intercepted.

[0034] The GPU detection method provided in this embodiment can timely discover GPUs with potential over-temperature risks by setting the deviation ratio threshold, and intercept these GPUs with potential over-temperature risks to ensure the stability and reliability of the system.

[0035] In an optional embodiment, the method further includes:

[0036] Based on the working environment temperature, determining a target deviation ratio threshold;

[0037] Among them, the deviation ratio thresholds corresponding to different working environment temperatures are different.

[0038] The graphics processor detection method provided in this embodiment can make the detection result of the graphics processor determined based on the deviation ratio threshold more accurate by setting different deviation ratio thresholds according to different working environment temperatures, thereby reducing false alarms or missed alarms of graphics processors with potential over-temperature risks.

[0039] In an optional embodiment, the method further includes:

[0040] Saving the test values ​​of the performance indicators of all the graphics processors to be tested and the detection results of the graphics processors to be tested;

[0041] Analyze all the stored test values ​​of the performance indicators of the graphics processor to be tested and the detection results of the graphics processor to be tested to obtain analysis results;

[0042] If the analysis result is that for the graphics processor to be tested in the same slot and at the same operating environment temperature, when the current deviation ratio threshold is used to detect the graphics processor, the number of intercepted graphics processors exceeds the preset number threshold, an alarm is issued to allow the user to adjust the current deviation ratio threshold.

[0043] The graphics processor detection method provided in this embodiment dynamically adjusts the deviation ratio threshold by storing the test values ​​of the performance indicators of all graphics processors to be tested and the detection results of the graphics processors to be tested, thereby reducing false positives or false negatives of graphics processors with potential over-temperature risks, continuously optimizing the detection process of the graphics processors, and improving the detection accuracy of the graphics processors.

[0044] In a second aspect, the present invention provides a graphics processor detection device, the device comprising:

[0045] The first pressurization module is used to perform full pressure pressurization processing on the graphics processor to be tested in the server to be tested by using the graphics processor pressurization tool during the server aging process at the factory end;

[0046] A first acquisition module, configured to perform a performance test on the graphics processor to be tested, and obtain a test value of a performance indicator of the graphics processor to be tested;

[0047] A second acquisition module is used to acquire the working environment temperature of the graphics processor to be tested;

[0048] A first screening module, configured to screen out a reference value of a performance indicator of a graphics processor corresponding to the working environment temperature from a target mapping table, wherein the target mapping table includes a correspondence between different working environment temperatures and reference values ​​of performance indicators of a graphics processor;

[0049] A first determination module, configured to determine a deviation ratio based on a test value and a reference value of the performance indicator;

[0050] The second determination module is used to determine that the graphics processor to be tested has a risk and fails the test when the deviation ratio exceeds a target deviation ratio threshold.

[0051] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the graphics processor detection method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0052] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to cause a computer to execute the graphics processor detection method of the first aspect or any corresponding embodiment thereof.

[0053] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the graphics processor detection method of the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related technologies, the drawings required for use in the specific embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 It is a flow chart of detecting whether a GPU has a potential over-temperature risk in the related art;

[0056] Figure 2 A schematic diagram of a flow chart of a method for detecting a graphics processor according to an embodiment of the present invention;

[0057] Figure 3 is a flowchart of another graphics processor detection method according to an embodiment of the present invention;

[0058] Figure 4 is a schematic diagram of a process of generating multiple mapping tables according to an embodiment of the present invention;

[0059] Figure 5 is a flowchart of another graphics processor detection method according to an embodiment of the present invention;

[0060] Figure 6 is a structural block diagram of a graphics processor detection device according to an embodiment of the present invention;

[0061] Figure 7 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0063] With the rapid development of high-performance computing, artificial intelligence, deep learning, and graphics rendering, graphics processors play a vital role in various computing tasks due to their powerful parallel computing capabilities and efficient data processing speed. They provide support for various computing-intensive tasks and are widely used in data centers, workstations, and personal computers.

[0064] However, the high performance of the GPU also brings about the problem of high temperature. Especially under high load conditions, the temperature of the GPU tends to rise rapidly. In related technologies, when the temperature of the GPU reaches or exceeds the preset safety threshold, the system automatically reduces the operating frequency of the GPU to reduce heat generation, protect the hardware, and prevent hardware damage. This mechanism causes GPU performance to decline, work efficiency to decrease, and user experience to decrease. Moreover, this mechanism will only sound an alarm when the temperature exceeds the threshold, which may have caused a certain degree of damage to the hardware.

[0065] In order to solve the above problems, relevant technologies intercept GPUs with potential overheating risks before AI servers leave the factory to prevent them from entering actual use. Figure 1 This is a flowchart of detecting whether a GPU has a potential over-temperature risk in the related technology, such as Figure 1 As shown, the related technology places the AI ​​server to be tested in a high-temperature aging chamber, powers on the AI ​​server to be tested, and controls the ambient temperature of the high-temperature aging chamber to 38 degrees Celsius to simulate the maximum operating temperature of the AI ​​server. After the AI ​​server to be tested is at an ambient temperature of 38 degrees Celsius for half an hour, the GPU pressure tool is run to perform full-stress pressure testing on the GPU of the AI ​​server to be tested, and the BMC of the AI ​​server to be tested is used to monitor the temperature of the GPU and the surrounding high-bandwidth memory (High Bandwidth Memory, referred to as: HBM) in real time. When the BMC monitors that the temperature of the GPU exceeds the temperature threshold, an alarm is triggered and the over-temperature alarm information is recorded. It should be noted that when the GPU is seriously over-temperature, the GPU will automatically power off, and then the AI ​​server will trigger a shutdown. At this time, the BMC will record the over-temperature shutdown information. After the full-stress pressure test of the GPU is completed, run the detection script to query whether there is an over-temperature alarm information or over-temperature shutdown information recorded in the BMC to determine whether the GPU has a potential over-temperature risk, and to detect whether the GPU has a potential over-temperature risk.

[0066] It should be noted that an AI server is a server with multiple GPU cards installed. GPUs with potential overheating risks can also be intercepted before any server containing GPUs leaves the factory. In the case of a potential overheating risk of a GPU, the BMC triggers an alarm to notify workers to repair it.

[0067] The maximum operating temperature of the AI ​​server can be 35 degrees Celsius to 38 degrees Celsius, and the above description uses 38 degrees Celsius as an example.

[0068] The related technology requires the use of an aging chamber that can control the ambient temperature to 38 degrees Celsius to detect whether the GPU has a potential over-temperature risk. This is expensive, and at an ambient temperature of 38 degrees Celsius, according to the AI ​​server's cooling strategy, its fan runs at full speed, consuming a lot of power, which increases electricity costs.

[0069] And it relies on a single temperature threshold to determine whether the GPU has a potential over-temperature risk. It is very likely that the GPU temperature will briefly exceed the temperature threshold, but there is no real over-temperature risk, resulting in low accuracy of GPU detection results. And the alarm will only be issued when the temperature exceeds the threshold, which may have caused a certain degree of damage to the hardware and reduced system performance. This detection method focuses on handling problems after they occur, and cannot perform preventive maintenance. For example, cooling measures are taken only when the temperature is too high, and potential risks cannot be discovered in time before that, which reduces the security of the server.

[0070] The embodiment of the present invention provides a graphics processor detection method. In the server aging link at the factory end, a graphics processor pressurization tool is used to perform full pressure pressurization processing on a graphics processor to be tested in the server to be tested, and a performance test is performed on the graphics processor to be tested, and a test value of a performance indicator of the graphics processor to be tested is obtained. The working environment temperature of the graphics processor to be tested is obtained, and a reference value of the performance indicator of the graphics processor corresponding to the working environment temperature is screened out from a target mapping table. A deviation ratio is determined according to the test value and the reference value of the performance indicator, and then a detection result of the graphics processor to be tested is determined according to a comparison between the deviation ratio and a target deviation ratio threshold value, so as to reduce costs, improve the accuracy of GPU detection results, and achieve an effect of earlier fault detection.

[0071] According to an embodiment of the present invention, an embodiment of a graphics processor detection method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0072] In this embodiment, a method for detecting a graphics processor is provided, which can be used in a mobile terminal, such as a server, a central processing unit, etc. Figure 2 is a flowchart of a method for detecting a graphics processor according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0073] Step S201, in the server aging stage at the factory end, a graphics processor stressing tool is used to perform full pressure stressing on the graphics processor to be tested in the server to be tested.

[0074] Among them, in order to intercept GPUs with potential over-temperature risks before AI servers leave the factory, so that GPUs entering the market will not have over-temperature risks, during the server aging stage at the factory, the GPUs in the servers to be tested are tested for potential over-temperature risks, so as to intercept GPUs with potential over-temperature risks.

[0075] It should be noted that the server to be tested can be an AI server or any server including a GPU.

[0076] First, it is necessary to use the graphics processor stressing tool in the relevant technology to fully stress the graphics processor to be tested in the AI ​​server to verify the performance of the GPU under high load.

[0077] Step S202: Perform a performance test on the graphics processor to be tested to obtain a test value of a performance indicator of the graphics processor to be tested.

[0078] After the graphics processor to be tested is subjected to full pressure processing, a performance test is performed on the graphics processor to be tested using a GPU performance test tool in the related art to obtain a test value of a performance indicator of the graphics processor to be tested.

[0079] It should be noted that the performance test of the graphics processor to be tested is performed to obtain the test values ​​of the performance indicators of the graphics processor to be tested, including:

[0080] A graphics processor performance test tool is used to perform multiple performance tests on the graphics processor to be tested, and multiple performance test results are obtained.

[0081] The multiple performance test results are averaged to obtain the test value of the performance index of the graphics processor to be tested.

[0082] The number of times the performance test is performed is set by a technician. For example, the performance test may be repeated three times to obtain the performance test results corresponding to the three performance tests, and the average of the performance test results corresponding to the three performance tests is taken to obtain the test value of the performance indicator of the graphics processor to be tested.

[0083] Step S203, obtaining the working environment temperature of the graphics processor to be tested.

[0084] Step S204: Filter out the reference value of the performance indicator of the graphics processor corresponding to the working environment temperature from the target mapping table. The target mapping table includes the correspondence between different working environment temperatures and the reference value of the performance indicator of the graphics processor.

[0085] After the working environment temperature of the graphics processor to be tested is obtained, a reference value of the performance indicator of the graphics processor corresponding to the working environment temperature is screened out from the target mapping table according to the working environment temperature of the graphics processor to be tested.

[0086] Step S205, determining a deviation ratio based on the test value and the reference value of the performance indicator.

[0087] After obtaining the reference value of the performance indicator of the graphics processor corresponding to the working environment temperature, the deviation ratio is determined according to the test value of the performance indicator of the graphics processor to be tested at the working environment temperature and the reference value of the performance indicator of the graphics processor corresponding to the working environment temperature.

[0088] Step S206: When the deviation ratio exceeds the target deviation ratio threshold, it is determined that the GPU to be tested has a risk and the test fails.

[0089] After obtaining the deviation ratio, it is determined whether the deviation ratio exceeds a target deviation ratio threshold. If the deviation ratio exceeds the target deviation ratio threshold, it is determined that the detection result of the graphics processor to be tested is that there is a risk, that is, a potential over-temperature risk, and the test fails.

[0090] It should be noted that the target deviation ratio threshold includes a positive deviation ratio threshold and a negative deviation ratio threshold. When the deviation ratio is a positive number, it is determined whether the deviation ratio is greater than the positive deviation ratio threshold. If the deviation ratio is greater than the positive deviation ratio threshold, it is determined that the deviation ratio exceeds the target deviation ratio threshold.

[0091] When the deviation ratio is a negative number, it is determined whether the deviation ratio is less than the negative deviation ratio threshold. If the deviation ratio is less than the negative deviation ratio threshold, it is determined that the deviation ratio exceeds the target deviation ratio threshold.

[0092] The numerical values ​​of the negative deviation ratio threshold and the positive deviation ratio threshold may be the same or different, and may be flexibly adjusted by technical personnel.

[0093] The graphics processor detection method provided in the embodiment uses a graphics processor pressurization tool to perform full pressure pressurization on the graphics processor to be tested in the server to be tested during the server aging stage at the factory end, performs a performance test on the graphics processor to be tested, obtains a test value of the performance index of the graphics processor to be tested, obtains the working environment temperature of the graphics processor to be tested, screens out a reference value of the performance index of the graphics processor corresponding to the working environment temperature from a target mapping table, determines a deviation ratio based on the test value and the reference value of the performance index, and then determines a detection result of the graphics processor to be tested based on a comparison between the deviation ratio and a target deviation ratio threshold, thereby improving the accuracy of the detection result and being able to issue an early warning when an abnormal situation has not occurred or has just occurred, thereby achieving earlier fault detection.

[0094] The graphics processor detection method provided in this embodiment obtains the performance test result of the graphics processor to be tested, that is, the test value of the performance index, under the room temperature environment of the factory end, determines the deviation ratio according to the test value of the performance index and the reference value, and then determines the detection result of the graphics processor to be tested according to the comparison between the deviation ratio and the target deviation ratio threshold value. The graphics processor can be tested without using a high-temperature aging chamber, thereby saving costs and reducing power consumption.

[0095] In this embodiment, a method for detecting a graphics processor is provided, which can be used in a mobile terminal, such as a server, a central processing unit, etc. Figure 3 is a flowchart of a method for detecting a graphics processor according to an embodiment of the present invention. Figure 3 As shown, the process includes the following steps:

[0096] Step S301: In the server aging phase at the factory end, a graphics processor stress tool is used to perform full pressure stress treatment on the graphics processor to be tested in the server to be tested. Figure 2 Step S201 of the illustrated embodiment will not be described in detail here.

[0097] Step S302: Perform a performance test on the GPU to be tested to obtain a test value of the performance index of the GPU to be tested. Figure 2 Step S202 of the illustrated embodiment will not be described in detail here.

[0098] Step S303, obtaining the working environment temperature of the graphics processor to be tested.

[0099] Specifically, the above step S303 includes:

[0100] Step S3031: Based on the baseboard management controller, the air inlet sensor temperature of the server to be tested is obtained.

[0101] The air inlet sensor temperature of the AI ​​server is obtained through the BMC.

[0102] Step S3032: Determine the working environment temperature of the graphics processor to be tested based on the air inlet sensor temperature of the server to be tested.

[0103] The air inlet sensor temperature of the AI ​​server is the working environment temperature of the GPU to be tested.

[0104] At this time, the room temperature value corresponding to the AI ​​server, that is, the working environment temperature of the graphics processor to be tested and the test value of the performance index of the graphics processor to be tested are obtained.

[0105] Step S304: Filter out the reference value of the performance index of the graphics processor corresponding to the working environment temperature from the target mapping table. The target mapping table includes the correspondence between different working environment temperatures and the reference value of the performance index of the graphics processor. Figure 2 Step S204 of the illustrated embodiment will not be described in detail here.

[0106] Step S305: Determine the deviation ratio based on the test value and the reference value of the performance indicator. Figure 2 Step S205 of the illustrated embodiment will not be described in detail here.

[0107] Step S306: If the deviation ratio exceeds the target deviation ratio threshold, it is determined that the GPU to be tested has a risk and fails the test. Figure 2 Step S206 of the illustrated embodiment will not be described in detail here.

[0108] It can be understood that when the deviation ratio does not exceed the target deviation ratio threshold, it is determined that the graphics processor to be tested does not have any risk, that is, there is no potential over-temperature risk, and the test passes.

[0109] After obtaining the deviation ratio, it is determined whether the deviation ratio exceeds the target deviation ratio threshold. If the deviation ratio does not exceed the deviation ratio threshold, it is determined that the test result of the graphics processor to be tested is that there is no potential over-temperature risk for the graphics processor to be tested, and the test has passed and can be put into market use.

[0110] When the deviation ratio is a positive number, it is determined whether the deviation ratio is greater than the positive deviation ratio threshold. If the deviation ratio is not greater than the positive deviation ratio threshold, it is determined that the deviation ratio does not exceed the target deviation ratio threshold.

[0111] When the deviation ratio is a negative number, it is determined whether the deviation ratio is less than the negative deviation ratio threshold. If the deviation ratio is not less than the negative deviation ratio threshold, it is determined that the deviation ratio does not exceed the target deviation ratio threshold.

[0112] The graphics processor detection method provided in this embodiment obtains the air inlet sensor temperature of the server to be tested based on the baseboard management controller, and then determines the working environment temperature of the graphics processor to be tested, thereby realizing the detection of the graphics processor to be tested under room temperature environment at the factory end. The graphics processor can be detected without using a high-temperature aging chamber, saving costs and reducing power consumption.

[0113] In some optional implementations, the above step S305 includes:

[0114] Step a1, determine the deviation ratio by the following formula:

[0115]

[0116] After obtaining the test value and the reference value of the performance indicator, the formula is used to determine the deviation ratio between the test value and the reference value.

[0117] The graphics processor detection method provided in this embodiment determines the deviation ratio according to the test value and the reference value of the performance indicator, and then determines the detection result of the graphics processor according to the comparison between the deviation ratio and the target deviation ratio threshold, thereby improving the accuracy of the detection result.

[0118] In some optional implementations, before filtering out the reference value of the performance indicator of the graphics processor corresponding to the working environment temperature from the target mapping table, the graphics processor detection method further includes:

[0119] Step b1, placing a plurality of normal servers in a constant temperature box, and setting the working environment temperature of the constant temperature box to a first temperature.

[0120] Before filtering out the reference value of the performance indicator of the graphics processor corresponding to the working environment temperature from the target mapping table, it is necessary to first determine the target mapping table.

[0121] Figure 4 FIG. 1 is a flow chart of generating multiple mapping tables according to an embodiment of the present invention. Figure 4 As shown, taking three normal servers as test prototypes and obtaining multiple mapping tables as an example, firstly, three normal servers are selected, placed in a constant temperature box, and powered on, and the working environment temperature of the constant temperature box is set to a first temperature, which is initially 5 degrees Celsius.

[0122] It should be noted that more or less normal servers than three normal servers may be used as test prototypes to obtain multiple mapping tables, and no specific limitation is made here. The normal servers and the server to be tested belong to the same type of server products.

[0123] Step b2: when the duration of the working environment temperature of the thermostat at the first temperature reaches a duration threshold, a graphics processor pressurization tool is used to perform full pressure pressurization processing on the graphics processors in the plurality of normal servers.

[0124] Among them, the duration threshold can be half an hour, that is, 0.5 hours. The duration that the working environment temperature of the constant temperature box is at the first temperature reaches the duration threshold, which means that the duration that the working environment temperature of the three normal servers is at the first temperature reaches the duration threshold. Then, the graphics processor pressurization tool is used to perform full pressure pressurization on the graphics processors in the three normal servers.

[0125] It should be noted that the GPU pressure tool used here is the same as the GPU pressure tool used in the aforementioned step S301.

[0126] Step b3: for any normal server, multiple performance tests are performed on the graphics processors in different slots of the normal server to obtain multiple performance test results of the graphics processors in different slots.

[0127] After completing the full-stress pressurization processing of the graphics processors of the three normal servers, for any normal server, multiple performance tests are performed on the graphics processors in different slots of the normal server using a GPU performance testing tool to obtain multiple performance test results of the graphics processors in different slots of the normal server.

[0128] Exemplarily, the multiple performance tests may be five performance tests. That is, for a graphics processor in any slot of a normal server, five performance tests are performed continuously to obtain five performance test results of the graphics processor in the slot of the normal server, and then five performance test results of the graphics processor in each slot of the normal server are obtained.

[0129] It should be noted that the GPU performance testing tool used here is the same as the GPU performance testing tool used in the aforementioned step S302.

[0130] Step b4, averaging multiple performance test results of graphics processors in the same slot in multiple normal servers to obtain benchmark values ​​of performance indicators of graphics processors in different slots.

[0131] After obtaining five performance test results of the graphics processor in each slot of the three normal servers, multiple performance test results (15 performance test results in this embodiment) of the graphics processor in the same slot of the three normal servers are averaged to obtain benchmark values ​​of performance indicators of graphics processors in different slots.

[0132] It should be noted that the slots of each normal server are configured the same, and the slot numbers of the slots at the same position in different normal servers are the same. The multiple performance test results of the graphics processors at the same slot in multiple normal servers are averaged, that is, the multiple performance test results of the graphics processors at the slots with the same slot numbers in multiple normal servers are averaged to obtain the benchmark value of the performance index of the graphics processor at the slot corresponding to each slot number.

[0133] It is understandable that when averaging the performance test results of multiple GPUs in the same slot in multiple normal servers, the weighted average value needs to be calculated after the deviation value is removed. In this embodiment, the weight of each GPU is the same.

[0134] Step b5, determining the next temperature to be measured based on the order of the temperature values ​​in the temperature range to be measured.

[0135] After obtaining the reference values ​​of the performance indicators of the graphics processors in different slots at 5 degrees Celsius, the next temperature to be measured is determined according to the order of the temperature values ​​in the temperature range to be measured.

[0136] The temperature range to be measured is from 5 degrees Celsius to 38 degrees Celsius. According to the order of temperature values ​​from low to high, the next temperature to be measured is determined by increasing the temperature by 1 degree Celsius each time.

[0137] For example, when the current working environment temperature of the thermostat is set to 5 degrees Celsius, the next temperature to be measured is 6 degrees Celsius.

[0138] Step b6, determining the next temperature to be measured as the first temperature, and returning to the step of setting the working environment temperature of the constant temperature box to the first temperature, until the test of all temperature values ​​in the temperature range to be measured is completed, and the benchmark values ​​of the performance indicators of the graphics processors in different slots under different working environment temperatures are obtained.

[0139] Among them, after the next temperature to be measured is determined, the next temperature to be measured is determined as the first temperature, and the process returns to execute the step of setting the working environment temperature of the constant temperature box to the first temperature and the subsequent steps b2 to b6, until the test of all temperature values ​​in the temperature range to be measured is completed, and the benchmark values ​​of the performance indicators of the graphics processors in different slots at different working environment temperatures are obtained, that is, all the test data of the three test prototypes are summarized, and the weighted average value is calculated after eliminating the deviation value, which is used as the benchmark value of the performance indicator.

[0140] It is understandable that after the performance test is completed, the test resources are released.

[0141] Step b7, determining a mapping table corresponding to the graphics processor in each slot based on the benchmark values ​​of the performance indicators of the graphics processors in different slots at different working environment temperatures.

[0142] After obtaining the benchmark values ​​of the performance indicators of the graphics processors in different slots at different working environment temperatures, the benchmark values ​​of the performance indicators of the graphics processors in different slots at different working environment temperatures are divided according to the slots, and a mapping table corresponding to the graphics processor in each slot is obtained, and the mapping table includes the corresponding relationship between different working environment temperatures and the benchmark values ​​of the performance indicators of the graphics processors.

[0143] Step b8: based on the slot where the GPU to be tested is located, select a target mapping table from the multiple mapping tables.

[0144] Among them, after obtaining the working environment temperature of the graphics processor to be tested, it is also necessary to obtain the slot where the graphics processor to be tested is located, that is, the slot number of the graphics processor to be tested, and according to the slot number of the graphics processor to be tested, select a target mapping table that matches the slot number of the graphics processor to be tested from multiple mapping tables.

[0145] The multiple mapping tables include the mapping table corresponding to the graphics processor of each slot obtained above.

[0146] The graphics processor detection method provided in this embodiment places multiple normal servers in a constant temperature box and sets different working environment temperatures, performs full pressure pressurization treatment and performance testing on the graphics processors of the multiple normal servers, obtains benchmark values ​​of performance indicators of graphics processors in different slots under different working environment temperatures, ensures the accuracy of the benchmark values ​​of the performance indicators obtained, and then detects whether the graphics processor to be tested has a potential over-temperature risk based on the obtained benchmark values ​​of the performance indicators, thereby improving the accuracy of the detection results of the graphics processor.

[0147] In some optional implementations, the graphics processor detection method further includes:

[0148] Step c1: when the deviation ratio exceeds the target deviation ratio threshold, an alarm is triggered and the graphics processor to be tested is intercepted.

[0149] Among them, after obtaining the deviation ratio, it is determined whether the deviation ratio exceeds the target deviation ratio threshold. If the deviation ratio exceeds the target deviation ratio threshold, an alarm is triggered and the graphics processor under test is intercepted, that is, the AI ​​server under test is intercepted and waits for workers to repair it.

[0150] It should be noted that the alarm is triggered by printing the alarm information on the display.

[0151] Regardless of whether the GPU test passes or fails, a test log is generated and uploaded to the production test system for storage. The test log includes the test item name, test data results of all GPUs in the AI ​​server, performance deviation values, etc.

[0152] For example, the name of the test item is GPU overtemperature detection. The test data results of all GPUs in the AI ​​server are test values ​​of performance indicators of all GPUs in the AI ​​server, and the performance deviation value is the deviation ratio.

[0153] At the same time, the test log can display the slot number of the GPU that failed the test in red font. The test log also includes the test results, that is, the test results of each GPU to be tested.

[0154] While waiting for workers to repair, all tests on the AI ​​server are stopped, and the AI ​​server sends commands to various internal high-speed serial communication bus (peripheral component interconnect express, referred to as: PCIE) devices to enter low-power mode, trying to keep the server running at a low power consumption and save electricity.

[0155] The GPU detection method provided in this embodiment can timely discover GPUs with potential over-temperature risks by setting the deviation ratio threshold, and intercept these GPUs with potential over-temperature risks to ensure the stability and reliability of the system.

[0156] In some optional implementations, the graphics processor detection method further includes:

[0157] Step d1, determining a target deviation ratio threshold based on the working environment temperature.

[0158] After obtaining the working environment temperature of the GPU to be tested, a target deviation ratio threshold value matching the working environment temperature of the GPU to be tested is selected from a mapping table of working environment temperature and deviation ratio threshold value according to the working environment temperature of the GPU to be tested.

[0159] The mapping table of the working environment temperature and the deviation ratio threshold is preset by a technician.

[0160] It should be noted that technicians can also make adaptive modifications to the mapping table of working environment temperature and deviation ratio threshold according to other environmental factors. For example, for products to be shipped to high altitude areas, it is necessary to strictly intercept GPUs with overheating risks, so the deviation ratio threshold corresponding to different working environment temperatures can be set smaller. For products to be shipped to low altitude areas, the heat dissipation environment is better, and the deviation ratio threshold corresponding to different working environment temperatures can be set larger.

[0161] It should be further explained that different deviation ratio thresholds can be determined for different AI server shipment levels.

[0162] The graphics processor detection method provided in this embodiment can make the detection result of the graphics processor determined based on the deviation ratio threshold more accurate by setting different deviation ratio thresholds according to different working environment temperatures, thereby reducing false alarms or missed alarms of graphics processors with potential over-temperature risks.

[0163] In some optional implementations, the graphics processor detection method further includes:

[0164] Step e1, saving the test values ​​of the performance indicators of all the graphics processors to be tested and the detection results of the graphics processors to be tested.

[0165] After the detection of whether each graphics processor to be tested has a potential over-temperature risk is completed, the test value of the performance indicator of the graphics processor to be tested and the detection result of the graphics processor to be tested are uploaded to the production test system for storage.

[0166] It should be noted that the performance test data of the AI ​​server other than the graphics processor to be tested can also be uploaded to the production test system for storage.

[0167] Step e2: analyzing the stored test values ​​of the performance indicators of all the graphics processors to be tested and the detection results of the graphics processors to be tested to obtain analysis results.

[0168] After a period of accumulation, a large amount of performance test data, namely, test values ​​of performance indicators of the graphics processor to be tested and test results of the graphics processor to be tested, is saved in the production test system. The saved data is analyzed to obtain analysis results.

[0169] Step e3: If the analysis result shows that for the graphics processor to be tested in the same slot and at the same operating environment temperature, when the current deviation ratio threshold is used to detect the graphics processor, the number of intercepted graphics processors exceeds the preset number threshold, an alarm is issued to allow the user to adjust the current deviation ratio threshold.

[0170] Among them, for the GPU to be tested in the same slot and the same working environment temperature, when the number of intercepted GPUs exceeds the preset number threshold when the current deviation ratio threshold in the slot and the working environment temperature is used to detect the GPU, an alarm is issued to allow the user to adjust the current deviation ratio threshold. The preset number threshold is set by the technician based on experience.

[0171] It is understandable that the strict deviation ratio threshold means that even sub-standard qualified products are intercepted, resulting in a large number of intercepted GPUs. Although this can improve product robustness, it also reduces the overall machine output efficiency.

[0172] A loose deviation ratio threshold will intercept fewer GPUs, which can improve overall output efficiency, but may allow some GPUs with potential over-temperature risks to enter the market.

[0173] Therefore, it is necessary to set an appropriate deviation ratio threshold according to different requirements.

[0174] If the number of intercepted GPUs exceeds the preset number threshold, it means that the currently set deviation ratio threshold is too strict and needs to be adjusted. The user can perform Gaussian distribution processing on the test values ​​of the performance indicators of all GPUs to be tested in the same slot and the same working environment temperature, and filter out the range of the test values ​​of the performance indicators of 99% of the GPUs to be tested in the slot and the working environment temperature, and then infer a more appropriate deviation ratio threshold for the slot and the working environment temperature.

[0175] The graphics processor detection method provided in this embodiment dynamically adjusts the deviation ratio threshold by storing the test values ​​of the performance indicators of all graphics processors to be tested and the detection results of the graphics processors to be tested, thereby reducing false positives or false negatives of graphics processors with potential over-temperature risks, continuously optimizing the detection process of the graphics processors, and improving the detection accuracy of the graphics processors.

[0176] In some optional implementations, the GPU has multiple performance indicators. A GPU performance testing tool is used to perform performance testing on the GPU to obtain test values ​​of multiple performance indicators. For any performance indicator, a deviation ratio of the performance indicator is obtained based on the test value and a benchmark value of the performance indicator.

[0177] The above step S306 may include:

[0178] Step f1: when the number of performance indicators whose deviation ratio exceeds the target deviation ratio threshold exceeds half of the total number of performance indicators, it is determined that the GPU to be tested has a potential over-temperature risk and the test fails.

[0179] It can be understood that when the number of performance indicators whose deviation ratio exceeds the target deviation ratio threshold does not exceed half of the total number of performance indicators, it is determined that there is no potential over-temperature risk for the graphics processor to be tested and the test passes.

[0180] The graphics processor detection method provided in this embodiment can more accurately evaluate the overall performance of the GPU, improve the detection accuracy of the GPU, and reduce the possibility of misjudgment by counting the number of performance indicators whose deviation ratio exceeds the target deviation ratio threshold and comparing it with the total number of performance indicators.

[0181] In this embodiment, a method for detecting a graphics processor is also provided. Figure 5 is a flowchart of a method for detecting a graphics processor according to an embodiment of the present invention. Figure 5 As shown, the method comprises the following steps:

[0182] The first step is to fully pressurize the graphics processor under test.

[0183] The second step is to perform performance testing on the graphics processor to be tested.

[0184] The third step is to import the target mapping table, which is the target mapping table in the aforementioned step S204.

[0185] The fourth step is to compare the test values ​​and benchmark values ​​of the collected performance indicators in real time to detect whether there are any abnormalities.

[0186] Specifically, the test value of the performance index of the graphics processor to be tested and the reference value of the performance index determine the deviation ratio, and then determine whether the graphics processor to be tested is abnormal, that is, whether the graphics processor to be tested has a potential over-temperature risk.

[0187] It should be noted that when determining whether the GPU to be tested is abnormal based on the deviation ratio, it is necessary to first determine a suitable deviation ratio threshold.

[0188] Step 5: When an abnormality is detected in the graphics processor to be tested, an alarm is triggered.

[0189] In the sixth step, interception operations are performed according to the preset strategy and the operation results are recorded. The GPU with an error is subsequently repaired by workers.

[0190] Among them, for the GPU under test with abnormality, the interception operation is performed according to the preset strategy, and the operation result is recorded, and the operation result is whether the interception is successful. The GPU with an error is the abnormal GPU and is subsequently repaired by the worker.

[0191] The preset strategy may be to intercept when anomalies occur.

[0192] Step 7: Save the performance test data and optimize the threshold setting and interception strategy based on the big data. Please refer to the description of the above steps e1 to e3 for details, which will not be repeated here.

[0193] The graphics processor detection method provided in this embodiment determines the deviation ratio according to the test value and the reference value of the performance indicator, and then determines the detection result of the graphics processor to be tested according to the comparison between the deviation ratio and the target deviation ratio threshold, thereby improving the accuracy of the detection result and being able to issue an early warning when an abnormal situation has not occurred or has just occurred, thereby achieving earlier fault detection.

[0194] In order to make the GPU detection method provided by this embodiment clearer, it is described in detail in conjunction with the following specific embodiments.

[0195] There are 8 GPU cards installed in the AI ​​server, and the 8 GPUs are numbered GPU0-GPU7 according to the physical slots. It can be understood that the slot numbers corresponding to the physical slots correspond to the GPU numbers of the GPUs one by one.

[0196] Use 3 normal AI servers and use steps b1 to b6 to test the benchmark values ​​of the performance indicators corresponding to each GPU card. Specifically, after each GPU is fully pressurized in the constant temperature box, each AI server uses the GPU performance test tool to test the performance test results of 8 GPUs one by one. After averaging the test values ​​of the performance indicators of the GPUs in the same slot of the 3 AI servers, the benchmark values ​​of the performance indicators of the GPUs corresponding to each slot at each working environment temperature are obtained. At this point, the benchmark values ​​of the performance indicators of the GPUs corresponding to different slots at different working environment temperatures are obtained, which are subsequently imported into the production test software for use.

[0197] The deviation ratio threshold needs to be set before production testing. For example, for orders sent to high altitudes, the positive deviation ratio threshold can be set to 5%, and the negative deviation ratio threshold can be set to -10%. For orders sent to low altitudes, the positive deviation ratio threshold can be set to 15%, and the negative deviation ratio threshold can be set to -15%. For orders with worse heat dissipation in the working environment, the positive deviation ratio threshold can be set to 5%, and the negative deviation ratio threshold can be set to -5%. At this time, the interception standard is more stringent, and more non-premium servers may be intercepted.

[0198] In the AI ​​server aging phase at the factory, the GPU under test in the server under test is fully pressurized, and the GPU performance test tool is called to obtain the test value of the performance index of the GPU under test. The air inlet sensor temperature of the AI ​​server, that is, the working environment temperature, is obtained through the BMC. At this time, the corresponding working environment temperature of the AI ​​server is obtained, that is, the working environment temperature of the GPU and the test value of the GPU's performance index.

[0199] According to the working environment temperature of the GPU and the slot where the GPU to be tested is located, the reference value of the performance index corresponding to the GPU in the slot under the working environment temperature is found, and the deviation ratio is determined according to the reference value and the test value of the performance index.

[0200] The deviation ratio is compared with the corresponding target deviation ratio threshold. When it is greater than the target deviation ratio threshold, it is judged that the GPU has a potential over-temperature risk, that is, there is a heat dissipation risk. Otherwise, it is judged that the GPU has no heat dissipation risk. When the GPU has a potential over-temperature risk, an alarm is triggered and the GPU is repaired or replaced.

[0201] The test values ​​and judgment results of the performance indicators of all GPUs in the order are saved in the cloud server test system at the factory end for retention, which is convenient for production test quality analysis and optimization of the deviation ratio threshold. If the deviation ratio threshold is set to ±5%, that is, the positive deviation ratio threshold is 5% and the negative deviation ratio threshold is -5%, the intercepted non-optimal GPU card ratio reaches 10%; when the deviation ratio threshold is set to ±10%, the intercepted non-optimal GPU card ratio reaches 0.5%; when the deviation ratio threshold is set to ±15%, the intercepted non-optimal GPU card ratio reaches 0.2%. Among them, the failure rate of GPU original factory-guided heat dissipation defective products is about 0.2%. At this time, it can be clearly seen that when the deviation ratio threshold is set to ±5%, more good GPU cards should be intercepted. At this time, the production test system script detects that too many GPU cards in the order are judged as heat dissipation defective products and are intercepted, indicating that the production test quality of this order is abnormal, triggering the emergency warning mechanism, and prompting that R&D and maintenance need to participate in analysis and processing.

[0202] The graphics processor detection method provided in this embodiment obtains the performance test results of the graphics processor to be tested, that is, the test value of the performance index, at room temperature in the factory, determines the deviation ratio according to the test value of the performance index and the reference value, and then determines the detection result of the graphics processor to be tested according to the comparison between the deviation ratio and the target deviation ratio threshold value, and can realize the detection of the graphics processor without using a high-temperature aging chamber, saving costs and reducing power consumption. According to the obtained reference value of the performance index, the graphics processor to be tested is tested for potential over-temperature risks, thereby improving the accuracy of the detection result of the graphics processor.

[0203] In this embodiment, a graphics processor detection system is also provided, which includes: a performance indicator collection module, a baseline establishment module, an anomaly detection module, a deviation ratio threshold setting module, an alarm and interception module, and a feedback and optimization module.

[0204] Among them, the performance indicator acquisition module is used to perform performance testing on the graphics processor.

[0205] The baseline establishment module is used to execute the above steps b1 to b6.

[0206] The anomaly detection module is used to execute the above steps S203 to S206.

[0207] The deviation ratio threshold setting module is used to set different deviation ratio thresholds according to different working environment temperatures.

[0208] The alarm and interception module is used to execute the above step c1.

[0209] The feedback and optimization module is used to execute the above steps e1 to e3.

[0210] The system is used to execute the graphics processor detection method shown in the above embodiment. For details, please refer to the description of the above corresponding steps, which will not be repeated here.

[0211] The graphics processor detection system provided in the present embodiment performs full pressure pressurization processing on the graphics processor to be tested in the server to be tested by using a graphics processor pressurization tool, performs performance testing on the graphics processor to be tested, obtains the test value of the performance index of the graphics processor to be tested, obtains the working environment temperature of the graphics processor to be tested, screens out the reference value of the performance index of the graphics processor corresponding to the working environment temperature from the target mapping table, determines the deviation ratio according to the test value of the performance index and the reference value, and then determines the detection result of the graphics processor to be tested according to the comparison between the deviation ratio and the target deviation ratio threshold value, thereby improving the accuracy of the detection result, and being able to issue an early warning when an abnormal situation has not occurred or has just occurred, thereby realizing earlier fault detection.

[0212] In this embodiment, a graphics processor detection device is also provided, which is used to implement the above embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0213] This embodiment provides a graphics processor detection device, such as Figure 6 As shown, including:

[0214] The first pressurization module 601 is used to perform full pressure pressurization processing on the graphics processor to be tested in the server to be tested by using a graphics processor pressurization tool during the server aging process at the factory end.

[0215] The first acquisition module 602 is used to perform a performance test on the GPU to be tested and obtain a test value of a performance indicator of the GPU to be tested.

[0216] The second acquisition module 603 is used to acquire the working environment temperature of the graphics processor to be tested.

[0217] The first screening module 604 is used to screen out the reference value of the performance indicator of the graphics processor corresponding to the working environment temperature from the target mapping table. The target mapping table includes the correspondence between different working environment temperatures and the reference value of the performance indicator of the graphics processor.

[0218] The first determination module 605 is used to determine the deviation ratio based on the test value and the reference value of the performance indicator.

[0219] The second determination module 606 is used to determine that the GPU to be tested has a risk and fails the test when the deviation ratio exceeds the target deviation ratio threshold.

[0220] In some optional implementations, the second acquisition module 603 includes:

[0221] The first acquisition unit is used to acquire the air inlet sensor temperature of the server to be tested based on the baseboard management controller.

[0222] The first determining unit is used to determine the working environment temperature of the graphics processor to be tested based on the air inlet sensor temperature of the server to be tested.

[0223] In some optional implementations, the graphics processor detection device further includes:

[0224] The setting module is used to place a plurality of normal servers in a constant temperature box and set the working environment temperature of the constant temperature box to a first temperature.

[0225] A second pressurizing module is used to perform full pressure pressurization processing on the graphics processors in the plurality of normal servers using a graphics processor pressurization tool when the duration of the working environment temperature of the thermostat box being at the first temperature reaches a duration threshold;

[0226] The third acquisition module is used to perform multiple performance tests on the graphics processors in different slots of any normal server to obtain multiple performance test results of the graphics processors in different slots.

[0227] The fourth acquisition module is used to average multiple performance test results of the graphics processors in the same slot in multiple normal servers to obtain benchmark values ​​of performance indicators of graphics processors in different slots.

[0228] The third determination module is used to determine the next temperature to be measured based on the high and low order of the temperature values ​​in the temperature range to be measured.

[0229] The fifth acquisition module is used to determine the next temperature to be measured as the first temperature, and return to execute the step of setting the working environment temperature of the constant temperature box to the first temperature until the test of all temperature values ​​in the temperature range to be measured is completed, so as to obtain the benchmark values ​​of the performance indicators of the graphics processors in different slots under different working environment temperatures.

[0230] The fourth determination module is used to determine a mapping table corresponding to the graphics processor in each slot based on the reference values ​​of the performance indicators of the graphics processors in different slots at different working environment temperatures.

[0231] The second screening module is used to screen out a target mapping table from the plurality of mapping tables based on the slot where the graphics processor to be tested is located.

[0232] In some optional implementations, the first determining module 605 includes:

[0233] The second determining unit is used to determine the deviation ratio by the following formula:

[0234]

[0235] In some optional implementations, the graphics processor detection device further includes:

[0236] The interception module is used to trigger an alarm and intercept the graphics processor to be tested when the deviation ratio exceeds the target deviation ratio threshold.

[0237] In some optional implementations, the graphics processor detection device further includes:

[0238] The fifth determination module is used to determine a target deviation ratio threshold based on the working environment temperature.

[0239] Among them, the deviation ratio thresholds corresponding to different working environment temperatures are different.

[0240] In some optional implementations, the graphics processor detection device further includes:

[0241] The saving module is used to save the test values ​​of the performance indicators of all the graphics processors to be tested and the detection results of the graphics processors to be tested.

[0242] The sixth acquisition module is used to analyze the stored test values ​​of the performance indicators of all the graphics processors to be tested and the detection results of the graphics processors to be tested to obtain analysis results.

[0243] The alarm module is used to issue an alarm if the analysis result shows that for the graphics processor to be tested in the same slot and at the same working environment temperature, when the number of intercepted graphics processors exceeds a preset number threshold when the current deviation ratio threshold is used for graphics processor detection, so that the user can adjust the current deviation ratio threshold.

[0244] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0245] The graphics processor detection device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0246] The embodiment of the present invention also provides a computer device having the above Figure 6 The graphics processor detection device shown.

[0247] See also Figure 7 , Figure 7 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 7 As shown, the computer device includes: one or more processors 701, memory 702, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Each component utilizes different buses to communicate with each other, and can be installed on a common mainboard or installed in other ways as required. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 7 A processor 701 is taken as an example.

[0248] The processor 701 may be a central processing unit, a network processor or a combination thereof. The processor 701 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable logic gate array, a general purpose array logic or any combination thereof.

[0249] The memory 702 stores instructions executable by at least one processor 701, so that the at least one processor 701 executes the method shown in the above embodiment.

[0250] The memory 702 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 702 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 702 may optionally include a memory remotely arranged relative to the processor 701, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0251] The memory 702 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 702 may also include a combination of the above types of memory.

[0252] The computer device also includes a communication interface 703, which is used for the computer device to communicate with other devices or a communication network.

[0253] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0254] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0255] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for detecting a graphics processor, characterized in that: The method comprises: In the server aging phase at the factory end, the GPU stressing tool is used to fully stress the GPU under test in the server under test; Performing a performance test on the graphics processor to be tested to obtain a test value of a performance indicator of the graphics processor to be tested; Obtaining the working environment temperature of the graphics processor to be tested; Filtering out a reference value of a performance indicator of a graphics processor corresponding to the working environment temperature from a target mapping table, wherein the target mapping table includes a correspondence between different working environment temperatures and reference values ​​of performance indicators of a graphics processor; Determining a deviation ratio based on the test value and the benchmark value of the performance indicator; When the deviation ratio exceeds the target deviation ratio threshold, it is determined that the graphics processor to be tested has a risk and the test fails.

2. The method according to claim 1, characterized in that The step of obtaining the working environment temperature of the graphics processor to be tested includes: Based on the baseboard management controller, obtaining the air inlet sensor temperature of the server to be tested; Based on the air inlet sensor temperature of the server to be tested, the operating environment temperature of the graphics processor to be tested is determined.

3. The method according to claim 1, characterized in that Before filtering out the reference value of the performance indicator of the graphics processor corresponding to the working environment temperature from the target mapping table, the method further includes: Placing a plurality of normal servers in a constant temperature box, and setting the working environment temperature of the constant temperature box to a first temperature; When the duration of the working environment temperature of the thermostat at the first temperature reaches a duration threshold, using a graphics processor pressurization tool to perform full pressure pressurization processing on graphics processors in multiple normal servers; For any normal server, multiple performance tests are performed on the graphics processors in different slots of the normal server to obtain multiple performance test results of the graphics processors in different slots; The performance test results of multiple GPUs in the same slot in multiple normal servers are averaged to obtain the benchmark values ​​of the performance indicators of GPUs in different slots. Determine the next temperature to be measured based on the order of the temperature values ​​in the temperature range to be measured; The next temperature to be measured is determined as the first temperature, and the step of setting the working environment temperature of the thermostat to the first temperature is returned to execute until all temperature values ​​in the temperature range to be measured are tested, and the benchmark values ​​of the performance indicators of the graphics processors in different slots under different working environment temperatures are obtained; Determine a mapping table corresponding to the graphics processor in each slot based on the benchmark values ​​of the performance indicators of the graphics processors in different slots at different working environment temperatures; The target mapping table is selected from the plurality of mapping tables based on the slot where the graphics processor to be tested is located.

4. The method according to claim 1, characterized in that: The step of determining the deviation ratio based on the test value and the reference value of the performance indicator comprises: The deviation ratio is determined by the following formula:

5. The method according to claim 1, characterized in that The method further comprises: When the deviation ratio exceeds a target deviation ratio threshold, an alarm is triggered and the graphics processor to be tested is intercepted.

6. The method according to claim 1, characterized in that The method further comprises: Based on the working environment temperature, determining a target deviation ratio threshold; Among them, the deviation ratio thresholds corresponding to different working environment temperatures are different.

7. The method according to claim 1, characterized in that The method further comprises: Saving the test values ​​of the performance indicators of all the graphics processors to be tested and the detection results of the graphics processors to be tested; Analyze all the stored test values ​​of the performance indicators of the graphics processor to be tested and the detection results of the graphics processor to be tested to obtain analysis results; If the analysis result is that for the graphics processor to be tested in the same slot and at the same operating environment temperature, when the current deviation ratio threshold is used to detect the graphics processor, the number of intercepted graphics processors exceeds the preset number threshold, an alarm is issued to allow the user to adjust the current deviation ratio threshold.

8. A graphics processor detection device, characterized in that: The device comprises: The first pressurization module is used to perform full pressure pressurization processing on the graphics processor to be tested in the server to be tested by using the graphics processor pressurization tool during the server aging process at the factory end; A first acquisition module, configured to perform a performance test on the graphics processor to be tested, and obtain a test value of a performance indicator of the graphics processor to be tested; A second acquisition module is used to acquire the working environment temperature of the graphics processor to be tested; A first screening module, configured to screen out a reference value of a performance indicator of a graphics processor corresponding to the working environment temperature from a target mapping table, wherein the target mapping table includes a correspondence between different working environment temperatures and reference values ​​of performance indicators of a graphics processor; A first determination module, configured to determine a deviation ratio based on a test value and a reference value of the performance indicator; The second determination module is used to determine that the graphics processor to be tested has a risk and fails the test when the deviation ratio exceeds a target deviation ratio threshold.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the graphics processor detection method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the graphics processor detection method according to any one of claims 1 to 7.

Images