Performance detection method, device, system and equipment of graphics processor, and medium

Abstract

The invention provides a performance detection method, device, system and equipment for a graphics processor and a medium, and relates to the technical field of computers, and the method comprises the steps: obtaining the performance test data of a current test graphics processor, and a performance interception standard corresponding to the model of the current test graphics processor; wherein the performance interception standard is determined based on the historical performance test data and the model of the current test graphics processor; determining a performance standard interval corresponding to the model of the current test graphics processor based on the performance interception standard; and determining a performance detection result of the current test graphics processor based on the performance test data of the current test graphics processor and the performance standard interval. Therefore, the performance detection of the graphics processor can be automatically realized, and the accuracy of the performance detection result of the graphics processor is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technologies, and in particular, to a method, apparatus, system, device, and medium for detecting the performance of a graphics processing unit. Background Art

[0002] With the continuous development of Artificial Intelligence (AI) technology, the demand for AI servers has increased, and the production volume of AI servers has also increased accordingly. Users' requirements for the quality and performance of the GPU (Graphics Processing Unit) components of AI servers have also become higher. In related technologies, it is usually only possible to perform hardware detection on the GPU to ensure that there are no hardware problems with the GPU, but it is not possible to perform performance detection on the GPU. Therefore, there is an urgent need for a method that can detect the performance of a graphics processing unit. Summary of the Invention

[0003] The present disclosure provides a method, apparatus, system, device, and medium for detecting the performance of a graphics processing unit. Its main purpose is to achieve the performance detection of a graphics processing unit.

[0004] According to a first aspect of the present disclosure, there is provided a method for detecting the performance of a graphics processing unit, including: Obtaining performance test data of a currently tested graphics processing unit, and a performance interception standard corresponding to the model of the currently tested graphics processing unit; wherein, the performance interception standard is determined based on the historical performance test data and model of the currently tested graphics processing unit; Determining a performance standard interval corresponding to the model of the currently tested graphics processing unit based on the performance interception standard; Determining a performance detection result of the currently tested graphics processing unit based on the performance test data of the currently tested graphics processing unit and the performance standard interval.

[0005] According to a second aspect of the present disclosure, there is provided a device for detecting the performance of a graphics processing unit, including: A data acquisition module, configured to obtain performance test data of a currently tested graphics processing unit, and a performance interception standard corresponding to the model of the currently tested graphics processing unit; wherein, the performance interception standard is determined based on the historical performance test data and model of the currently tested graphics processing unit; An interval determination module, configured to determine a performance standard interval corresponding to the model of the currently tested graphics processing unit based on the performance interception standard; A performance detection module, configured to determine a performance detection result of the currently tested graphics processing unit based on the performance test data of the currently tested graphics processing unit and the performance standard interval.

[0006] According to a third aspect of the present disclosure, a performance detection system for a graphics processor is provided, including a test machine and a server side, wherein: The test machine is configured to: Obtain performance test data of the currently tested graphics processor, and a performance interception criterion corresponding to the model of the currently tested graphics processor; wherein, the performance interception criterion is determined based on historical performance test data and the model of the currently tested graphics processor; Determine a performance standard interval corresponding to the model of the currently tested graphics processor based on the performance interception criterion; Determine a performance detection result of the currently tested graphics processor based on the performance test data of the currently tested graphics processor and the performance standard interval; The server side is configured to update the performance interception criterion corresponding to the model of the currently tested graphics processor based on the performance test data determined to be valid data.

[0007] According to a fourth aspect of the present disclosure, an electronic device is provided, including: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the method described in the foregoing first aspect.

[0008] According to a fifth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method described in the foregoing first aspect.

[0009] According to a sixth aspect of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the method described in the foregoing first aspect is implemented.

[0010] Through the present disclosure, by obtaining performance test data of a current tested graphics processing unit and a performance interception standard corresponding to the model of the current tested graphics processing unit, wherein the performance interception standard is determined based on historical performance test data and the model of the current tested graphics processing unit; determining a performance standard range corresponding to the model of the current tested graphics processing unit based on the performance interception standard; and determining a performance detection result of the current tested graphics processing unit based on the performance test data of the current tested graphics processing unit and the performance standard range. In this way, performance detection of the graphics processing unit can be achieved, and since the performance interception standard is determined based on historical performance test data and the model of the tested graphics processing unit, the accuracy of the performance detection result of the graphics processing unit can also be improved.

[0011] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. Among them: Figure 1 is a schematic flowchart of a method for detecting the performance of a graphics processing unit provided by an embodiment of the present disclosure; Figure 2 is a schematic diagram of a configuration file provided by an embodiment of the present disclosure; Figure 3 is a schematic structural diagram of a system for detecting the performance of a graphics processing unit provided by an embodiment of the present disclosure; Figure 4 is a schematic diagram of the operation logic of a test machine end provided by an embodiment of the present disclosure; Figure 5 is a schematic diagram of the format of an NCCL log provided by an embodiment of the present disclosure; Figure 6 is a schematic diagram of the execution logic of a server end provided by an embodiment of the present disclosure; Figure 7 is a schematic diagram of the recording format of a GPU model and performance test data provided by an embodiment of the present disclosure; Figure 8 is a schematic structural diagram of a device for detecting the performance of a graphics processing unit provided by an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] The exemplary embodiments of the present disclosure will be described below in conjunction with the accompanying drawings. Various details of the embodiments of the present disclosure are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, descriptions of well-known functions and structures are omitted in the following description for clarity and conciseness.

[0014] As can be seen from the background art, in the related art, usually only the shipped GPUs can be guaranteed to have no hardware problems. The performance of the GPU is very important for user services. Insufficient GPU performance may affect the user service results. For example, if a GPU with insufficient performance is used for large model training, the training results may be inaccurate. Therefore, it is necessary to perform performance testing on the GPU before leaving the factory, which requires a performance standard to determine whether the tested GPU performance is qualified. The NCCL (NVIDIA Collective Communications Library) tool library is mainly used for efficient communication between multiple GPUs. The ALL_reduce tool is a tool in the NCCL library, and its test results can reflect the communication rate between GPUs, and thus reflect the performance status of the GPU.

[0015] However, NVIDIA does not provide a clear GPU performance standard. It is only possible to collect a large amount of data for comparative analysis to determine a relatively appropriate performance standard that can ensure that most GPUs meet this standard, while intercepting individual GPUs with lower performance to prevent them from flowing into the client. For the performance standard, if it is set too low, it may cause GPUs with substandard performance to flow to users, affecting the normal business of users and causing direct or indirect losses; if the performance standard is set too high, a large number of GPUs will be judged as unqualified and cannot be processed, which will also cause losses to the manufacturer. In a similar scenario, if some standard files need to be adjusted, it is usually necessary to collect, classify, and screen historical data, and then calculate a new standard based on a large number of historical data. However, if there are individual abnormal data in the collected data, the calculated standard will become inaccurate, and even affect the subsequent data accuracy and product quality.

[0016] Based on this, the present disclosure provides a method, device, system, equipment, and medium for detecting the performance of a graphics processor, which can collect and process test data in real time, eliminate the interference of abnormal data, and ensure the accuracy of the interception standard.

[0017] The method, device, and electronic equipment for detecting the performance of a graphics processor according to the embodiments of the present disclosure will be described below with reference to the accompanying drawings.

[0018] Figure 1Flow chart of a method for detecting the performance of a graphics processor provided by an embodiment of the present disclosure.

[0019] As Figure 1 shown, the method includes the following steps: Step 101, obtain performance test data of the currently tested graphics processor and a performance interception standard corresponding to the model of the currently tested graphics processor.

[0020] Among them, the performance interception standard is determined based on historical performance test data and models of test graphics processors with the same model as the currently tested graphics processor.

[0021] Among them, the currently tested graphics processor may be the graphics processing unit (GPU) currently being tested. Obtaining performance test data of the currently tested graphics processor and a performance interception standard corresponding to the model of the currently tested graphics processor may be obtained in the format of a configuration file. For example, the configuration file may be defined in the Figure 2 shown format. For example, it may be named nccl.cfg. The first column in the configuration file may be GPU model information (GPU Model), the second column may be the current performance interception standard for this model of GPU (i.e., the NCCL test interception standard) (standard), and the third column may be the number of test machines on which this model of GPU has undergone NCLL tests (allcount). As an example, the model of the currently tested graphics processor can be determined first, and then the performance interception standard corresponding to this model can be obtained through the configuration file.

[0022] Among them, the performance interception standards corresponding to different models of test graphics processors may be determined in advance based on historical performance test data of different models of test graphics processors. That is, different models of test graphics processors may correspond to different performance interception standards.

[0023] Step 102, determine a performance standard range corresponding to the model of the currently tested graphics processor based on the performance interception standard.

[0024] Among them, the performance standard range may be a data range used to measure whether the performance of the currently tested graphics processor meets the standard (meets the requirements). The performance standard range is similar to a 'cut-off score' used to judge the performance level of the currently tested graphics processor. The performance standard range may be calculated based on the aforementioned obtained performance interception standard corresponding to the model of the currently tested graphics processor.

[0025] Step 103, determine the performance detection result of the currently tested graphics processor based on the performance test data and the performance standard range of the currently tested graphics processor.

[0026] Among them, the performance test data of the currently tested graphics processing unit can be matched with the performance standard range to determine whether the performance test data of the currently tested graphics processing unit falls within the performance standard range. If the performance test data of the currently tested graphics processing unit falls within the performance standard range, it can be considered that the performance detection result of the currently tested graphics processing unit is qualified in performance, that is, the performance meets the requirements; conversely, if the performance test data of the currently tested graphics processing unit does not fall within the performance standard range, it can be considered that the performance detection result of the currently tested graphics processing unit is unqualified in performance, that is, the performance does not meet the requirements.

[0027] In summary, the method provided by the embodiments of the present disclosure obtains the performance test data of the currently tested graphics processing unit and the performance interception standard corresponding to the model of the currently tested graphics processing unit; wherein, the performance interception standard is determined based on the historical performance test data and the model of the currently tested graphics processing unit; the performance standard range corresponding to the model of the currently tested graphics processing unit is determined based on the performance interception standard; based on the performance test data of the currently tested graphics processing unit and the performance standard range, the performance detection result of the currently tested graphics processing unit is determined. In this way, the performance detection of the graphics processing unit can be realized, and since the performance interception standard is determined based on the historical performance test data and the model of the tested graphics processing unit, the accuracy of the performance detection result of the graphics processing unit can also be improved.

[0028] It should be noted that there may be multiple steps in the embodiments of the present disclosure. For the convenience of description, these steps are numbered, but these numbers are not intended to limit the execution time slots and execution orders between the steps; these steps can be implemented in any order, and the embodiments of the present disclosure do not make any limitations in this regard.

[0029] In some embodiments, when determining the performance standard range corresponding to the model of the currently tested graphics processing unit based on the performance interception standard, the following steps can be adopted: Determine the lower limit of the performance standard range corresponding to the model of the currently tested graphics processing unit based on the first preset floating ratio and the performance interception standard; Determine the first upper limit of the performance standard range corresponding to the model of the currently tested graphics processing unit based on the second preset floating ratio and the performance interception standard; Determine the performance standard range based on the lower limit and the first upper limit.

[0030] Among them, the first preset floating ratio can be used to determine the lower limit of the performance standard range. For example, the first preset floating ratio can be 1%. The second preset floating ratio can be used to determine the upper limit (the first upper limit) of the performance standard range. For example, the second preset floating ratio can be 2%. The lower limit of the performance standard range corresponding to the model of the currently tested graphics processing unit can be calculated based on the first preset floating ratio and the performance interception standard. For example, it can be a floating range of 1% lower than the current interception standard. The first upper limit of the performance standard range corresponding to the model of the currently tested graphics processing unit can be determined based on the second preset floating ratio and the performance interception standard. For example, it can be a floating range of 2% higher than the current interception standard. In this way, the accuracy of the performance standard range can be improved, and the accuracy of the detection result can be improved.

[0031] In some embodiments, when determining the performance detection result of the currently tested graphics processing unit based on the performance test data and the performance standard range of the currently tested graphics processing unit, the following steps can be adopted: If the performance test data of the currently tested graphics processing unit is less than the lower limit of the performance standard range, it is determined that the performance detection result of the currently tested graphics processing unit does not meet the standard, and the performance test data is determined to be invalid data; If the performance test data of the currently tested graphics processing unit is greater than or equal to the lower limit of the performance standard range, then it is determined whether the performance test data of the currently tested graphics processing unit is greater than the first upper limit of the performance standard range; If the performance test data of the currently tested graphics processing unit is greater than the first upper limit of the performance standard range, it is determined that the performance detection result of the currently tested graphics processing unit meets the standard, and the performance test data is determined to be invalid data.

[0032] Among them, the performance test data of the tested graphics processor can be compared with the lower limit of the performance standard range to determine whether the performance test data of the currently tested graphics processor is less than the lower limit of the performance standard range. If the performance test data of the currently tested graphics processor is less than the lower limit of the performance standard range, it can be considered that the performance detection result of the currently tested graphics processor fails to meet the standard, and the performance test data can be determined as invalid data and will not be used to calculate the new performance interception standard. As an example, assume that the performance interception standard is 65.7360, and floating down by 1% is 65.7360 - 65.7360 * 0.01 = 65.07864, that is, the lower limit of the performance standard range is 65.07864. Since the NCCL test result of the currently tested graphics processor can usually only be accurate to 4 decimal places, the lower limit of the performance standard range can be rounded to 65.0786. If the NCCL test result of the current test machine is lower than 65.0786, it is considered that the GPU NCCL performance of the current test machine fails to meet the standard, the test machine reports an error and requires further handling by the staff, and the NCCL data of the test machine is considered invalid data and will not be calculated into the new interception standard; if the NCCL test result of the current test machine is greater than 65.0786, then enter the process of determining whether the performance test data is greater than the first upper limit of the performance standard range below.

[0033] If the performance test data of the currently tested graphics processor is greater than the first upper limit of the performance standard range, it can be considered that the performance detection result of the currently tested graphics processor meets the standard, that is, it meets the performance requirements. As an example, assume that the performance interception standard is 65.7360, and floating up by 2% is 65.7360 + 65.7360 * 0.02 = 67.05072, which is 67.0507 after rounding. If the performance test data of the currently tested graphics processor is greater than 67.0507, it is determined that the NCCL test item passes and the performance of the currently tested graphics processor meets the standard. However, the determined data is invalid data, and the relatively high data of this test result will not be calculated into the new interception standard. Because the NCCL test data of the vast majority of GPUs of the same model will be concentrated in a relatively small range (conforming to the normal distribution law), generally the floating range will not exceed 1% upward or downward. Occasionally, the NCCL test results of individual test machines will be relatively high. Therefore, the purpose of intercepting the NCCL test results is to intercept the relatively low GPU NCCL performance of a small number of cases. If the data with an individual data greater than the 2% floating range is calculated into the interception standard, the interception standard will become higher, resulting in more test machines being determined to have low NCCL performance subsequently. In this way, by judging the test results through the performance standard range with high accuracy calculated above, the effectiveness and accuracy of the detection results can be further improved.

[0034] In some embodiments, when determining the performance detection result of the current test graphics processor based on the performance test data and the performance standard range of the current test graphics processor, the following steps may also be adopted: Determine whether the performance test data of the current test graphics processor is greater than or equal to the lower limit of the performance standard range and less than or equal to the second upper limit of the performance standard range; wherein, the second upper limit is determined based on the performance interception standard and the third preset floating ratio; If the performance test data of the current test graphics processor is greater than or equal to the lower limit of the performance standard range and less than or equal to the second upper limit of the performance standard range, determine that the performance detection result of the current test graphics processor is qualified; Determine the performance test data of the current test graphics processor as valid data.

[0035] Among them, the third preset floating ratio may be the upward floating range based on the performance interception standard, such as 1%. The second upper limit can be calculated based on the performance interception standard and the third preset floating ratio. Then, determine whether the performance test data of the current test graphics processor is greater than or equal to the lower limit of the performance standard range and less than or equal to the second upper limit of the performance standard range, that is, determine whether the performance test data of the current test graphics processor is within the range of the first preset floating ratio (such as 1%) downward floating and the third preset floating ratio (such as 1%) upward floating of the performance interception standard. If the performance test data of the current test graphics processor is greater than or equal to the lower limit of the performance standard range and less than or equal to the second upper limit of the performance standard range, it can be determined that the performance detection result of the current test graphics processor is qualified. That is, if the performance test data of the current test graphics processor is within the range of the first preset floating ratio (such as 1%) downward floating and the third preset floating ratio (such as 1%) upward floating of the performance interception standard, it can also be considered that the performance detection result of the current test graphics processor is qualified, and the performance test data of the current test graphics processor can be determined as valid data for calculating the new performance interception standard data. In this way, through the determination of the strict floating ratio range, it can be ensured that only qualified data is recorded and used, ensuring the accuracy and reliability of the test results; these valid data will be used to update the new performance interception standard, which can make the performance interception standard closer to the actual situation; the valid data is directly used for subsequent calculations, which can reduce the interference of invalid data and improve the efficiency of testing and standard updating.

[0036] In some embodiments, after determining the performance detection result of the current test graphics processor based on the performance test data and the performance standard range of the current test graphics processor, the following steps are further included: Update the performance interception standard corresponding to the model of the current test graphics processor based on the performance test data determined as valid data.

[0037] Among them, the performance test data determined to be valid data can be the performance test data within the range of floating downward by a first preset floating ratio (such as 1%) and floating upward by a third preset floating ratio (such as 1%) from the performance interception standard. In this way, by using the valid performance test data, the performance interception standard can be updated, so that the updated performance interception standard can better reflect the actual performance level of the current graphics processor, improving the accuracy of the performance interception standard; also, with the continuous addition of new valid data, continuous optimization of the performance interception standard can be achieved to ensure that subsequent tests more accurately evaluate the performance of the graphics processor and realize dynamic optimization of performance evaluation; at the same time, updating the performance interception standard based on valid data can also reduce misjudgments caused by data deviation, making the test results more credible and enhancing the reliability of the test.

[0038] In some embodiments, when updating the performance interception standard corresponding to the model of the currently tested graphics processor based on the performance test data determined to be valid data, the following steps can be adopted: Read the number of test graphics processors corresponding to the model of the currently tested graphics processor and the target performance test data corresponding to the test graphics processors of the model; Calculate a new performance interception standard based on the performance interception standard, the number of test graphics processors, and the target performance test data corresponding to the model of the currently tested graphics processor; Update the performance interception standard corresponding to the model of the currently tested graphics processor based on the new performance interception standard.

[0039] Among them, the target performance test data can be the performance test data determined to be valid data.

[0040] Among them, when updating the performance interception standard, the number of test graphics processors corresponding to the model of the currently tested graphics processor and the target performance test data corresponding to the test graphics processors of the model can be read first. Then, a new performance interception standard is calculated based on the performance interception standard, the number of test graphics processors, and the target performance test data corresponding to the model of the currently tested graphics processor. For example, the calculation method can be as follows:

[0041] Among them, the number of test machines for this model of GPU is the number of test graphics processors, the NCCL test data of the test machines recorded in the current file is the target performance test data recorded in the current configuration file, and the current interception standard is the current performance interception standard.

[0042] After calculating the new performance interception standard, the performance interception standard corresponding to the model of the currently tested graphics processing unit can be updated based on the new performance interception standard. For example, write the new performance interception standard into the configuration file nccl.cfg, increment the number of tested graphics processing units corresponding to the model by one, and delete the nccl_{sn}.cfg (serial number of the tested graphics processing unit in the configuration file) file that has just been traversed. In this way, accurate calculation of the new performance interception standard can be achieved, and the accuracy and rationality of the updated performance interception standard can be improved.

[0043] To make the performance detection method of the graphics processing unit provided by the embodiments of the present disclosure clearer, the following will be described with specific examples.

[0044] When the performance detection method of the graphics processing unit provided by the present disclosure is used for the first time, a test interception standard (performance interception standard) needs to be set according to the existing data. After the GPU server completes the NCCL test, it will automatically run the performance detection system of the graphics processing unit provided by the present disclosure, and dynamically adjust based on this standard according to the latest performance test data, so as to adjust the performance interception standard in real time according to the latest performance test data.

[0045] See Figure 3 , the performance detection method of the graphics processing unit provided by the present disclosure can be run in the performance detection system of the graphics processing unit. The performance detection system of the graphics processing unit includes a test machine and a server side, and the server side is a server that stores configuration files separately. On the test machine side (such as a tested graphics processing unit), new performance test data can be processed; on the server side, the update of the new performance interception standard is completed, and both the test machine side and the server side are carried out simultaneously to achieve real-time adjustment and update of the NCCL interception standard. Specifically as follows: The operation logic of the test machine side can be seen in Figure 4 , such as Figure 4 shown, it may include the following steps: 1. After the NCCL test is completed, the system provided by the present disclosure (performance detection system of the graphics processing unit) is automatically run, and the test machine side will capture the NCCL test data (performance test data) of the current test machine (current tested graphics processing unit) in the NCCL log. For example, the NCCL test result (performance test data) of the current test machine can be determined by capturing the keyword "Avg bus bandwidth (average bus bandwidth)", and the format of the NCCL log can be as Figure 5 shown, Figure 5The NCCL test result of the test machine is 65.2846; the GPU model of the current test machine can be obtained by capturing the "Product Name" in the nvidia-smi (NVIDIA System Management Interface) information.

[0046] 2. The format of the configuration file can be as Figure 2 shown and can be named nccl.cfg. Among them, the first column is the GPU model information, the second column is the current NCCL test interception standard (performance interception standard) for this model of GPU, and the third column is the number of test machines that have undergone NCCL tests for this model of GPU, that is, the current interception standard is calculated based on this number. The NCCL performance interception standard and the number of test machines for this GPU model can be determined through the GPU model.

[0047] 3. Calculate whether the NCCL test result of the current test machine is lower than the floating range of 1% of the current performance interception standard. Taking GPU HXX as an example, the current test interception standard is 65.7360, and the downward floating by 1% is 65.7360 - 65.7360 * 0.01 = 65.07864. Since the NCCL test data result can only be accurate to 4 decimal places, it is rounded to 65.0786 after rounding. If the NCCL test result of the current test machine is lower than 65.0786, it is considered that the GPU NCCL performance of the current test machine does not meet the standard, and the test machine reports an error and requires further processing by the staff. Also, the NCCL data of the test machine is considered invalid data and will not be calculated into the new interception standard; if the NCCL test result of the current test machine is greater than 65.0786, the following process is entered.

[0048] 4. Calculate whether the current NCCL test result of the test machine is within a 2% floating range higher than the current interception standard. Taking GPU HXX as an example again, the upper floating limit by 2% is 65.7360 + 65.7360 * 0.02 = 67.05072, which is rounded to 67.0507. If the NCCL test result of the current test machine is greater than 67.0507, it is determined that the NCCL test item passes, but the determined data is invalid data (because for the vast majority of NCCL test data of the same model of GPU, it will be concentrated within a relatively small range (conforming to the normal distribution law), generally the floating range will not exceed 1% upward or downward. Occasionally, there will be individual test machines with relatively high NCCL test results. The purpose of intercepting the NCCL test results is to intercept the relatively low NCCL performance of a small number of GPUs. If the data with an individual value greater than the 2% floating range is included in the interception standard, the interception standard will become higher, resulting in more test machines being determined to have low NCCL performance subsequently, which is not what we want to happen). The test results with relatively high values will not be calculated into the new interception standard; 5. If the NCCL test result of the current test machine is within the range of 1% downward and 1% upward, it is also determined that the NCCL test item passes, and this data is valid data ( Figure 5 The NCCL test results in the log shown in Figure 7 are valid data). Write the NCCL data of this test machine into nccl_{sn}.cfg. As

[0049] shown, it will record the GPU model and NCCL test data of the current test machine and upload them to the directory where the configuration file is stored on the server. This test data will be calculated into the new interception standard.

[0049] The execution logic on the server side can be referred to Figure 6 . When writing the new performance interception standard into the configuration file, there may be a problem of conflicts caused by multiple test machines reading and writing the configuration file simultaneously. To solve this problem, the following operations will be performed when updating the configuration file: 1. Traverse all files in the format of nccl_{sn}.cfg in the directory where the configuration file is stored and process them sequentially. As Figure 7 shown, read the GPU model information (the first column) and the NCCL test data of the test machine (the second column) in the file.

[0050] 2. Recalculate the new interception standard using (current interception standard * the number of test machines of this model of GPU + the NCCL test data of the test machine recorded in the current file) / (the number of test machines of this model of GPU + 1). Taking Figure 5 and Figure 7For example, the new interception standard = (65.7360 * 2480 + 65.2846) / (2480 + 1) = 65.7095.

[0051] 3. Write the new interception standard into nccl.cfg, increment the number of test machines for the corresponding GPU by one, and delete the nccl_{sn}.cfg file that has just been traversed. Repeat the above operations until there are no more nccl_{sn}.cfg files in the current directory.

[0052] In this way, at least the following technical effects can be achieved: 1. Ensure the quality of the produced GPU test machines and prevent low-performance test machines from flowing into users; 2. Can automatically adjust the performance interception standard in real time, reducing the workload of production line staff and ensuring the timeliness of the performance interception standard, thereby guaranteeing product quality; 3. When dynamically adjusting the NCCL performance interception standard, the number of test machines will be recorded in the configuration file, eliminating the problem of counting the production volume of a certain model of GPU; 4. Can make the NCCL performance interception standard more accurate and reasonable, avoiding the irrationality of manually setting the performance interception standard; 5. Can solve the problem of low accuracy in manually determining the performance interception standard and improve the accuracy of intercepting anomalies; 6. Can also solve the situation where multiple people may process the same file simultaneously in the case of manual processing, resulting in conflicts in modified content, and avoid abnormal situations.

[0053] According to an embodiment of the present disclosure, the present disclosure also provides a performance detection device for a graphics processor.

[0054] Exemplarily, Figure 8 FIG. 8 is a schematic structural diagram of a performance detection device for a graphics processor provided by an embodiment of the present disclosure. The performance detection device 800 for the graphics processor includes: a data acquisition module 801, an interval determination module 802, and a performance detection module 803; wherein, The data acquisition module 801 is configured to acquire performance test data of the currently tested graphics processor and a performance interception standard corresponding to the model of the currently tested graphics processor; wherein, the performance interception standard is determined based on historical performance test data and the model of the currently tested graphics processor; The interval determination module 802 is configured to determine a performance standard interval corresponding to the model of the currently tested graphics processor based on the performance interception standard; The performance detection module 803 is configured to determine a performance detection result of the currently tested graphics processor based on the performance test data of the currently tested graphics processor and the performance standard interval.

[0055] Further, the interval determination module 802 is specifically configured to: Determine the lower limit of the performance standard range corresponding to the model of the current tested graphics processing unit based on the first preset floating ratio and the performance interception standard; Determine the first upper limit of the performance standard range corresponding to the model of the current tested graphics processing unit based on the second preset floating ratio and the performance interception standard; Determine the performance standard range based on the lower limit and the first upper limit.

[0056] Further, the performance detection module 803 is specifically configured to: If the performance test data of the current tested graphics processing unit is less than the lower limit of the performance standard range, determine that the performance detection result of the current tested graphics processing unit is unqualified, and determine the performance test data as invalid data; If the performance test data of the current tested graphics processing unit is greater than or equal to the lower limit of the performance standard range, then check whether the performance test data of the current tested graphics processing unit is greater than the first upper limit of the performance standard range; If the performance test data of the current tested graphics processing unit is greater than the first upper limit of the performance standard range, determine that the performance detection result of the current tested graphics processing unit is qualified, and determine the performance test data as invalid data.

[0057] Further, the performance detection module 803 is specifically configured to: Determine whether the performance test data of the current tested graphics processing unit is greater than or equal to the lower limit of the performance standard range and less than or equal to the second upper limit of the performance standard range; wherein, the second upper limit is determined based on the performance interception standard and the third preset floating ratio; If the performance test data of the current tested graphics processing unit is greater than or equal to the lower limit of the performance standard range and less than or equal to the second upper limit of the performance standard range, determine that the performance detection result of the current tested graphics processing unit is qualified; Determine the performance test data of the current tested graphics processing unit as valid data.

[0058] Further, it further includes a standard update module, which is used to: Update the performance interception standard corresponding to the model of the current tested graphics processing unit based on the performance test data determined as valid data.

[0059] Further, the standard update module is specifically configured to: Read the number of tested graphics processing units corresponding to the model of the current tested graphics processing unit, and the target performance test data corresponding to the tested graphics processing units of the model; wherein, the target performance test data is the performance test data determined as valid data; Calculate a new performance interception criterion based on the performance interception criterion corresponding to the model of the current test graphics processing unit, the number of test graphics processing units, and the target performance test data; Update the performance interception criterion corresponding to the model of the current test graphics processing unit based on the new performance interception criterion.

[0060] It should be noted that for the description of the features in the embodiments corresponding to the graphics processing unit performance detection device, reference can be made to the relevant descriptions in the embodiments corresponding to the graphics processing unit performance detection method, which will not be elaborated here one by one.

[0061] An embodiment of the present disclosure also provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the embodiments of the above-mentioned graphics processing unit performance detection method.

[0062] An embodiment of the present disclosure also provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps in any one of the embodiments of the above-mentioned graphics processing unit performance detection method when running.

[0063] In an exemplary embodiment, the above-mentioned computer-readable storage medium may include, but is not limited to: USB flash drives, read-only memories (ROM for short), random access memories (RAM for short), mobile hard disks, magnetic disks, or optical discs, and other various media that can store computer programs.

[0064] An embodiment of the present disclosure also provides a computer program product. The above-mentioned computer program product includes a computer program, and when the computer program is executed by a processor, it implements the steps in any one of the embodiments of the above-mentioned graphics processing unit performance detection method.

[0065] An embodiment of the present disclosure also provides another computer program product, including a non-volatile computer-readable storage medium. The non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps in any one of the embodiments of the above-mentioned graphics processing unit performance detection method.

[0066] Those skilled in the art may further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present disclosure.

[0067] The above has introduced in detail a method for detecting the performance of a graphics processor provided by the present disclosure. Specific examples have been used herein to elaborate on the principles and implementation manners of the present disclosure. The description of the above embodiments is only used to help understand the method and its core idea of the present disclosure. It should be noted that for those of ordinary skill in the art in the technical field, without departing from the principles of the present disclosure, several improvements and modifications can be made to the present disclosure, and these improvements and modifications also fall within the protection scope of the claims of the present disclosure.

Claims

1. A method for detecting the performance of a graphics processor, characterized in that, Including: Obtaining performance test data of the current test graphics processing unit, and a performance interception standard corresponding to the model of the current test graphics processing unit; wherein, the performance interception standard is determined based on historical performance test data and models of test graphics processing units with the same model as the current test graphics processing unit; Determining a performance standard range corresponding to the model of the current test graphics processing unit based on the performance interception standard; Determining a performance detection result of the current test graphics processing unit based on the performance test data of the current test graphics processing unit and the performance standard range.

2. The method according to claim 1, wherein The determining a performance standard range corresponding to the model of the current test graphics processing unit based on the performance interception standard includes: Determining a lower limit of the performance standard range corresponding to the model of the current test graphics processing unit based on a first preset floating ratio and the performance interception standard; Determining a first upper limit of the performance standard range corresponding to the model of the current test graphics processing unit based on a second preset floating ratio and the performance interception standard; Determining the performance standard range based on the lower limit and the first upper limit.

3. The method according to claim 2, characterized in that, The determining a performance detection result of the current test graphics processing unit based on the performance test data of the current test graphics processing unit and the performance standard range includes: If the performance test data of the current test graphics processing unit is less than the lower limit of the performance standard range, determining that the performance detection result of the current test graphics processing unit fails to meet the standard, and determining the performance test data as invalid data; If the performance test data of the current test graphics processing unit is greater than or equal to the lower limit of the performance standard range, then determining whether the performance test data of the current test graphics processing unit is greater than the first upper limit of the performance standard range; If the performance test data of the current test graphics processing unit is greater than the first upper limit of the performance standard range, determining that the performance detection result of the current test graphics processing unit meets the standard, and determining the performance test data as invalid data.

4. The method according to claim 3, wherein The determining a performance detection result of the current test graphics processing unit based on the performance test data of the current test graphics processing unit and the performance standard range further includes: Determining whether the performance test data of the current test graphics processing unit is greater than or equal to the lower limit of the performance standard range and less than or equal to a second upper limit of the performance standard range; wherein, the second upper limit is determined based on the performance interception standard and a third preset floating ratio; If the performance test data of the current test graphics processing unit is greater than or equal to the lower limit of the performance standard range and less than or equal to the second upper limit of the performance standard range, determining that the performance detection result of the current test graphics processing unit meets the standard; Determining the performance test data of the current test graphics processing unit as valid data.

5. The method according to any one of claims 1-4, characterized in that, After the determining a performance detection result of the current test graphics processing unit based on the performance test data of the current test graphics processing unit and the performance standard range, further including: Updating the performance interception standard corresponding to the model of the current test graphics processing unit based on the performance test data determined as valid data.

6. The method according to claim 5, characterized in that, Updating the performance interception criteria corresponding to the model of the current test graphics processing unit based on the performance test data determined to be valid data includes: Reading the number of test graphics processing units corresponding to the model of the current test graphics processing unit and the target performance test data corresponding to the test graphics processing units of the model; wherein, the target performance test data is the performance test data determined to be valid data; Calculating new performance interception criteria based on the performance interception criteria corresponding to the model of the current test graphics processing unit, the number of test graphics processing units, and the target performance test data; Updating the performance interception criteria corresponding to the model of the current test graphics processing unit based on the new performance interception criteria.

7. A performance detection device for a graphics processor, characterized in that, Including: A data acquisition module, configured to acquire the performance test data of the current test graphics processing unit and the performance interception criteria corresponding to the model of the current test graphics processing unit; wherein, the performance interception criteria are determined based on the historical performance test data and the model of the current test graphics processing unit; An interval determination module, configured to determine the performance standard interval corresponding to the model of the current test graphics processing unit based on the performance interception criteria; A performance detection module, configured to determine the performance detection result of the current test graphics processing unit based on the performance test data of the current test graphics processing unit and the performance standard interval.

8. A performance detection system for a graphics processor, characterized in that, Including a test machine and a server side, wherein: The test machine is configured to: Acquire the performance test data of the current test graphics processing unit and the performance interception criteria corresponding to the model of the current test graphics processing unit; wherein, the performance interception criteria are determined based on the historical performance test data and the model of the current test graphics processing unit; Determine the performance standard interval corresponding to the model of the current test graphics processing unit based on the performance interception criteria; Determine the performance detection result of the current test graphics processing unit based on the performance test data of the current test graphics processing unit and the performance standard interval; The server side is configured to update the performance interception criteria corresponding to the model of the current test graphics processing unit based on the performance test data determined to be valid data.

9. An electronic device, characterized in that It includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the method according to any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, Stores computer instructions; wherein, the computer instructions are used to cause a computer to execute the method according to any one of claims 1 to 6.

Images