Software and hardware combined GPU (Graphics Processing Unit) automatic testing method and system

Abstract

The invention is suitable for the technical field of GPU testing, and provides a software and hardware combined GPU automatic testing method and system, which comprises a hardware platform and a software platform, and is characterized in that the software platform is used for setting a voltage source output voltage value, power-on time and power-off time of the hardware platform, obtaining current, voltage and temperature data fed back by the hardware platform, and sending the current, voltage and temperature data to the hardware platform; associating the data with the automatic test GPU performance, and establishing a GPU test fault case database; and the hardware platform is used for providing power for the GPU, controlling power-on and power-off of the GPU and a power supply voltage value of the GPU according to the setting of the software platform, reading real-time current, voltage and temperature data near a GPU pin and feeding back the real-time current, voltage and temperature data to the software platform. According to the method, automatic testing of the GPU is achieved in a software and hardware combination mode, data graphical analysis can be achieved, abnormal problems occurring in the testing process can be positioned by comparing analysis results, and the GPU detection efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of GPU testing, and in particular relates to a GPU automatic testing method and system combining software and hardware. Background Art

[0002] In recent years, domestic GPU software and hardware technologies have flourished, and a large number of domestic GPUs have emerged in the market. The domestic platform has formed a certain degree of autonomy and control, and the industry and ecology have gradually become sound. In view of the instability of display or hardware status of the current mainstream domestic graphics processors during production and operation, it is necessary to test each device equipped with a domestic graphics processor to ensure the quality of the hardware equipment. Testing is an indispensable part of multi-platform preparation based on domestic graphics processors. Executing test cases is an important task for testers in daily testing, and it is also one of the more complicated tasks. Therefore, for the multi-platform of domestic graphics processors, it is of great significance to have a complete and reliable automated testing method.

[0003] The existing domestic graphics processor automatic detection technology has the following main problems:

[0004] First of all, for the current mainstream domestic graphics processors, the automated detection technology lacks flexibility and versatility; for example, during the production, debugging, and testing stages, it is necessary to perform multiple types of automated tests on different architectures and different devices at the same time, and the existing technology cannot meet the requirements.

[0005] Secondly, functional testing and batch testing are cumbersome to operate. Most of them are performed manually by testers, who use simple simulation tools to send signals manually and judge the results by eye. This method is inefficient and time-consuming. There are thousands of test cases, and manual execution is prone to fatigue. Testers perform manual testing as needed, but the test coverage often fails to meet the requirements, which brings difficulties to the evaluation of GPU functions, performance, and applicability.

[0006] Third, during the testing process, most testing solutions only perform automated testing on the GPU's computing power and video memory pressure from a software perspective. If GPU function abnormalities or unstable performance are encountered during the testing process, there is no hardware data feedback, and there is no next debugging direction for the GPU function abnormalities that occur, making it difficult to locate the problem.

[0007] In addition, from the patent technology, there are currently some patent disclosures about GPU testing, such as the patent application "A method for automatically testing GPUs" disclosed by application number 201811237877.7. This application uses the open source testing tool NVQUAL for GPU testing. The application testing method is only applicable to NVIDIA graphics cards and does not have the versatility to deal with different model testing methods. For another example, the patent application "A systematic automatic testing method for GPUs" disclosed by application number 202211492477.7 first presets the test dimensions, builds corresponding test algorithms for the preset test dimensions, and finally analyzes the GPU data based on the test results. The entire process requires the construction of corresponding algorithms based on the tested system, and does not have the versatility for different test demand platforms. In addition, this invention patent is based on the construction of the algorithm and has poor versatility. Another example is the patent application "A multi-platform automated testing method and system based on domestic graphics processors" published with application number 202311512967.3. This method only systematically and automatically tests the GPU functions and performance from a software perspective. During the test process, there is no correlation with the GPU functions and performance during the test process from a hardware perspective. There is no hardware data support for abnormal situations that occur during the test process, and there is no certain guiding direction for solving abnormal situations.

[0008] The current existing technical solutions, whether they are actually used technologies or patented technologies, can only perform GPU function and performance test screening, and cannot provide data support and problem location for the next step of GPU debugging and GPU application. Summary of the invention

[0009] In view of the above problems, the purpose of the present invention is to provide a GPU automatic testing method and system combining software and hardware, aiming to solve the above technical problems.

[0010] The present invention adopts the following technical solution:

[0011] On the one hand, the hardware-software combined GPU automated testing system includes a hardware platform and a software platform.

[0012] The software platform is used to set the voltage source output voltage value, power-on time and power-off time of the hardware platform, and obtain the current, voltage and temperature data fed back by the hardware platform, and associate the data with the automated test GPU performance, establish a GPU test fault case database, analyze the specific parameters and influencing factors that affect the GPU function and performance, locate the abnormal problems that occur during the test, and realize GPU automated testing;

[0013] The hardware platform is used to provide GPU power supply, and control GPU power on and off and GPU power supply voltage value according to the settings of the software platform, and read the real-time current, voltage and temperature data near the GPU pins and feed them back to the software platform;

[0014] The hardware platform includes a voltage source, a voltage acquisition module, a current acquisition module, and a temperature reading module. The voltage acquisition module, the current acquisition module, and the temperature reading module are respectively connected to the GPU to be tested, and the voltage source supplies power to the GPU; the software platform is implemented through a console, and the console is connected to the voltage source and the voltage acquisition module, the current acquisition module, and the temperature reading module.

[0015] Furthermore, the console is a module or chip with communication interaction, display, and fast computing capabilities, and the console includes:

[0016] A voltage source regulating unit, used to realize digital control of the voltage source and perform real-time regulation of the voltage source;

[0017] A calculation and analysis unit, used to store, calculate and analyze the voltage data output by the voltage acquisition module, the current data output by the current acquisition module, and the GPU shell temperature data output by the temperature reading module;

[0018] The graphical unit is used to display and compare the output voltage value of the voltage source, the voltage data output by the voltage acquisition module, the current data output by the current acquisition module, and the temperature data output by the temperature reading module in real time, and draw corresponding curves to show the fluctuation range and change trend of the voltage, current, power consumption, and temperature when the GPU is working;

[0019] The test unit is used to perform performance testing on the GPU. During the test, the output voltage value of the voltage source, as well as the GPU pin voltage value, current value, temperature value, and power consumption are monitored in real time, and the test results are associated with the monitoring data.

[0020] Furthermore, the voltage source includes an external reference power supply, a precision unit and a voltage superposition circuit. The voltage source provides power supply voltage for the GPU. Multiple precision units are connected in series between the external reference power supply and the ground. The total resistance of each precision unit is fixed. Each precision unit is divided according to the resistance value. The precision unit controls the connected resistance value through a channel selector to change the connected power supply voltage value.

[0021] On the other hand, the GPU automated testing method combining software and hardware is implemented based on the GPU automated testing system, and the GPU automated testing method includes:

[0022] Step S1, GPU chip performance test: using the test case library and example execution instruction scripts to automatically test GPU computing and video memory pressure;

[0023] Step S2, data graphical processing: after the GPU chip performance test is completed, the test data is processed by using software tools, combined with the data fed back by the hardware platform, for graphical display and data graphical analysis;

[0024] Step S3, establishing a GPU test database: collecting and analyzing data on abnormal situations that occur during the test, and establishing a test failure case library.

[0025] Furthermore, the specific process of the GPU chip performance test is as follows:

[0026] S11. Design test items and build a complete set of test cases according to the localized GPU chip platform and localized GPU models, and deploy automated testing tools;

[0027] S12, after selecting the test items and test cases for the software test of the current GPU chip, set the test parameters, including the test interval, the number of tests, and the test running time;

[0028] S13. Generate test execution instructions according to the selected test items and test cases, as well as the set test parameters, and obtain the test case script file of the test instance from the specified directory;

[0029] S14. Execute the test case script file to perform automated testing, and display the current test status locally in real time. After the automated testing is completed, generate a test report with the test results and save the test data. At the same time, read the current, voltage, temperature, and power consumption data fed back by the hardware platform in real time during the test, and record the data change trend, minimum value, and maximum value.

[0030] Furthermore, the selected test cases include conventional performance tests and extreme performance tests, wherein the conventional performance tests include restart test cases, 2D performance test cases, 3D test cases, pixel fill rate test cases, texture fill rate test cases and computing power test cases, and the extreme performance tests include using a multi-process method to continuously increase the number of 2D test cases, 3D test cases and video playbacks, and at the same time determine whether the number of corresponding test case processes increases synchronously, to obtain the maximum number of 2D test cases, 3D test cases and video playbacks running simultaneously, and save the test cases and test results in excel files named after the current test dimension, and save them in the excel total file named after GPU.

[0031] Furthermore, the specific process of the data graphical processing is as follows:

[0032] Read the excel1 file and draw relevant data graphs based on the test data fed back by the hardware platform during the routine performance test and the extreme performance test to show the changes in current, voltage, temperature, and power consumption of the GPU during the running of different test cases. The data graphs are divided into two categories, namely, single test graphical display and X / Y axis comparison graphical display of multiple test cases.

[0033] Furthermore, the specific process of establishing the GPU test database is as follows:

[0034] During the extreme performance test, data on abnormal situations that occur during the test is collected and analyzed, and a GPU test fault case library is established to provide preliminary debugging directions for abnormal problems that may arise in subsequent GPU chip product applications and to locate the problems.

[0035] The beneficial effects of the present invention are as follows: the present invention provides a method and system for automatic GPU detection that combines software and hardware, which can realize automatic testing of multi-platform clients, including completing the testing of basic functions and various performances of GPU chips. The present invention combines software and hardware. During the test process, the console collects data fed back by the hardware and associates it with the performance displayed during the software testing process. The console processes the test data, performs graphical data analysis, collects and analyzes data on abnormal situations that occur during the test process, and establishes a test fault case library. The test fault case library uses a large amount of test data for comparative analysis. The comparative analysis results can locate abnormal problems that occur during the test process, provide data support and theoretical basis for the subsequent practical application process of the GPU, improve the efficiency of GPU detection, and ensure the subsequent normal application of the GPU. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a structural block diagram of a GPU automated testing system provided by an embodiment of the present invention;

[0037] Figure 2 is a connection structure diagram of a hardware platform and a software platform provided by an embodiment of the present invention;

[0038] Figure 3 is a structural diagram of a voltage source provided by an embodiment of the present invention;

[0039] Figure 4 It is a schematic diagram of the process of GPU chip performance testing provided by an embodiment of the present invention;

[0040] Figure 5 It is a schematic diagram of the process of data graphical processing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] Since there is currently no complete set of automated GPU automated detection systems that can be applied to all domestic GPUs, most testers perform manual tests as needed, and the test coverage often cannot meet the requirements. The test process is also relatively cumbersome and time-consuming, which brings difficulties to the evaluation of GPU functions, performance and applicability. At the same time, during the test process, most automated test schemes only perform automated testing and recording of GPU calculation and video memory pressure from a software perspective. If GPU function abnormalities and unstable performance are encountered during the test process, they are simply screened without hardware data feedback. Therefore, the current existing scheme can only perform GPU function and performance test screening, and cannot provide data support and problem direction positioning for the next step of GPU debugging and application. Therefore, in order to adapt to the requirements of different models of GPU test calculation and video memory, and provide test data support for hardware adaptation, debugging, and application in the GPU application process, the present invention designs a GPU automatic detection method and system that combines software and hardware. In order to realize automated testing of multi-platform clients, the test of basic functions and various performances of GPU chips is completed. In order to illustrate the technical solution described in the present invention, it is described below through specific embodiments.

[0043] like Figure 1 As shown, the GPU automatic testing system combining software and hardware provided in this embodiment includes a hardware platform and a software platform.

[0044] The software platform is used to set the voltage source output voltage value, power-on time and power-off time of the hardware platform, and obtain the current, voltage and temperature data fed back by the hardware platform, and associate the data with the automated test GPU performance, establish a GPU test fault case database, analyze the specific parameters and influencing factors that affect the GPU function and performance, locate the abnormal problems that occur during the test, and realize GPU automated testing;

[0045] The hardware platform is used to provide GPU power supply, and control GPU power on and off and GPU power supply voltage value according to the settings of the software platform, and read the real-time current, voltage and temperature data near the GPU pins and feed them back to the software platform.

[0046] Combination Figure 2As shown, the hardware platform includes a voltage source, a voltage acquisition module, a current acquisition module, and a temperature reading module. The voltage acquisition module, the current acquisition module, and the temperature reading module are respectively connected to the GPU to be tested, and the voltage source supplies power to the GPU; the software platform is implemented through a console, and the console is connected to the voltage source and the voltage acquisition module, the current acquisition module, and the temperature reading module.

[0047] In this system, the software platform is a console, which collects real-time current, voltage, and temperature data fed back by the hardware platform, and associates it with the GPU performance test during the software automated testing of the GPU chip. It establishes a GPU test fault case database, analyzes the specific parameters and influencing factors that affect the GPU function and performance, and locates abnormal problems that occur during the test process, providing data support and theoretical basis for the subsequent actual application process of the GPU, improving the GPU detection efficiency, and ensuring the normal application of the subsequent GPU.

[0048] The console is a module or chip with communication interaction, display, and fast computing capabilities, such as a single-chip microcomputer, SOC chip, host computer, etc. as a computing and control center. The main functions of the console include performing GPU-related calculations, software testing of video memory pressure, and controlling the voltage source to complete the reception, processing, analysis, and display of voltage, current, and temperature data. According to the specific function implementation, the console includes:

[0049] The voltage source regulating unit is used to realize digital control of the voltage source and perform real-time regulation of the voltage source. The voltage source regulating unit is mainly responsible for the communication control and operation part of the voltage source;

[0050] A calculation and analysis unit, used to store, calculate and analyze the voltage data output by the voltage acquisition module, the current data output by the current acquisition module, and the GPU shell temperature data output by the temperature reading module;

[0051] The graphical unit is used to display and compare the output voltage value of the voltage source, the voltage data output by the voltage acquisition module, the current data output by the current acquisition module, and the temperature data output by the temperature reading module in real time, and draw corresponding curves to show the fluctuation range and change trend of the voltage, current, power consumption, and temperature when the GPU is working;

[0052] The test unit is used to perform performance tests on the GPU, including GPU memory testing and computing stress testing. During the test, the output voltage value of the voltage source, as well as the GPU pin voltage value, current value, temperature value, and power consumption are monitored in real time. The test results are associated with the monitoring data to combine hardware feedback data with the GPU operating status.

[0053] For the structure of the voltage source, combined Figure 3As shown, the voltage source includes an external reference power supply, a precision unit and a voltage superposition circuit. The voltage source provides a power supply voltage for the GPU. A plurality of precision units (such as M (M≥1) precision units) are connected in series between the external reference power supply and the ground. The total resistance of each precision unit is fixed. Each precision unit is divided according to the resistance value. The precision unit controls the size of the connected resistance through a channel selector to change the connected power supply voltage value. Specifically, the precision unit is composed of a resistor array and a channel selector. The different resistance values ​​of the resistors in the resistor array can constitute precision units of different precisions. The precision units of different precisions can adjust the output of different step voltage values. The resistor array in the precision unit is composed of a plurality of resistors of the same resistance value and the same precision connected in series. Each resistor connection node is divided according to the resistance value and has a certain voltage value. Each resistor connection node of the resistor array is connected to the channel selector, and the number of channels of the channel selector is equal to the number of resistors. Through the selective conduction of the channel selector, the power supply voltage can be switched in the same resistor array, and different voltage values ​​can be selected for output. By selecting precision units with different accuracies, the voltage step accuracy can be switched, and the output power supply voltage with higher accuracy and smaller step value can be achieved, thereby realizing precise control of the GPU power supply voltage.

[0054] The voltage acquisition module consists of two main parts, namely the sensor and the analog-to-digital converter. The sensor part is responsible for measuring the voltage, usually using a high-precision resistor. This sensor is placed near the GPU power pin to ensure that the voltage changes of the GPU can be accurately captured. Through this layout design, the sensor can sense and transmit voltage information in real time, thereby improving the overall measurement accuracy and response speed. The analog-to-digital converter part converts the analog voltage signal measured at both ends of the sensor into an easy-to-process digital signal, making subsequent data processing and analysis more convenient and efficient. A high-performance analog-to-digital converter is selected to ensure that the collected data has high resolution and low error, thereby providing a reliable foundation for the system's voltage monitoring.

[0055] The current acquisition module is mainly composed of sensors and analog-to-digital converters. The sensor is responsible for measuring the current. Common sensor types include current transformers and Hall sensors. The analog-to-digital converter is a key component for converting the analog signal output by the sensor into a digital signal. The converted digital signal can be easily processed, analyzed and stored. The current acquisition module measures the current through the sensor and converts it into an easy-to-process signal, and then converts these signals into digital form through the analog-to-digital converter, providing high-precision and reliable current data for GPU testing.

[0056] For the temperature reading module, the main actions performed are temperature sensing, signal conversion, signal processing and digital display. First, temperature sensing uses a high-precision temperature sensor that can quickly and accurately sense changes in ambient temperature and convert these changes into corresponding electrical signals. Then, these electrical signals are processed by an analog-to-digital converter (ADC) to convert the analog signal output by the sensor into an easy-to-process digital signal. The converted digital signal enters the signal processing unit. Third, in signal processing, the signal will undergo further processing and correction to ensure the accuracy and reliability of the measurement results. Finally, the processed digital signal is transmitted to the digital display unit for intuitively displaying the current temperature reading. Through a series of steps such as temperature sensing, signal conversion, signal processing and digital display, the temperature reading module can accurately measure and display the temperature in the GPU working environment, provide accurate and reliable temperature data, and help users grasp the system status in real time and perform effective temperature management and control.

[0057] Based on the above system structure, this embodiment also provides a GPU automated testing method combining software and hardware. The GPU automated testing method includes three parts, which are:

[0058] Step S1, GPU chip performance test: using the test case library and example execution instruction scripts to automatically test GPU computing and video memory pressure;

[0059] Step S2, data graphical processing: after the GPU chip performance test is completed, the test data is processed by using software tools, combined with the data fed back by the hardware platform, for graphical display and data graphical analysis;

[0060] Step S3, establishing a GPU test database: collecting and analyzing data on abnormal situations that occur during the test, and establishing a test failure case library.

[0061] For the first part, GPU chip performance test, the specific process is as follows:

[0062] S11. Design test items and build a complete set of test cases according to the domestic GPU chip platform and domestic GPU model, and deploy automated testing tools.

[0063] Different models of GPU graphics cards may have different resources and cases tested during the automatic testing process, or the tester may have specific test focuses, so the test items can be designed by the tester during the automatic testing process. According to the installed domestic platform and the model of domestic GPU, the automated testing tool is deployed, and the corresponding test case set is matched.

[0064] S12, after selecting the test items and test cases for the software test of the current GPU chip, set the test parameters, including the test interval time, the number of tests, and the test running time.

[0065] The platform interface displays all test items and test cases of the automated testing software for testers to choose from. Testers can complete test items and test cases based on their actual needs and set relevant parameters, including test interval time, test number, test running time, etc.

[0066] S13. Generate test execution instructions according to the selected test items and test cases, as well as the set test parameters, and obtain the test case script file of the test instance from the specified directory.

[0067] Test case execution and test parameters are used to form test execution instructions. Combined with local platform information, the test case script file of the test instance is run and executed from the specified directory.

[0068] S14. Execute the test case script file to perform automated testing, and display the current test status locally in real time. After the automated testing is completed, generate a test report with the test results and save the test data. At the same time, read the current, voltage, temperature, and power consumption data fed back by the hardware platform in real time during the test, and record the data change trend, minimum value, and maximum value.

[0069] During the above GPU chip performance test, real-time GPU testing and data reading are performed, and the console stores the changes in current, voltage, and temperature fed back by the hardware platform, and records the change trends, minimum values, maximum values, and other hardware circuit parameters.

[0070] Combination Figure 4 As shown, the selected test cases include conventional performance tests and extreme performance tests, wherein the conventional performance tests include restart test cases, 2D performance test cases, 3D test cases, pixel fill rate test cases, texture fill rate test cases and computing power test cases, and the extreme performance test includes using a multi-process method to continuously increase the number of 2D test cases, 3D test cases and video playback, and at the same time determine whether the number of corresponding test case processes increases synchronously, to obtain the maximum number of 2D test cases, 3D test cases and video playback running simultaneously, and save the test cases and test results in excel files named after the current test dimension, and save them in the excel total file named after GPU.

[0071] For the above test case, here is a further explanation:

[0072] Restart test: repeatedly power on and off the GPU, set the restart times and restart interval;

[0073] 2D performance test: Run mainstream 2D ​​test cases or customized 2D test cases, set the total test run time; run the number and types of test cases at the same time.

[0074] 3D performance testing: Run mainstream 3D test cases or customized 3D test cases, set the total test run time; run the number and types of test cases at the same time.

[0075] Pixel fill rate test: Run mainstream pixel fill rate test cases or customized pixel fill rate test cases, set the total test run time; run the number and types of test cases at the same time.

[0076] Texture fill rate test: Run mainstream texture fill rate test cases or customized texture fill rate test cases, set the total test run time; run the number and types of test cases at the same time.

[0077] Computing power test: Run mainstream computing power test cases or customized computing power test cases, set the total test running time; run the number and types of test cases at the same time.

[0078] Test the maximum number of 2D test cases that can be run simultaneously: Run mainstream 2D ​​test cases or customized 2D test cases, set the time interval for adding test cases, and set the upper limit of the number of test cases that can be run simultaneously. Increase the number of test cases that can be run at a fixed interval. When the GPU is stuck or has excessive load, record the number of test cases that can be run and stop the test.

[0079] Test the maximum number of 3D test cases that can be run simultaneously: Run mainstream 3D test cases or customized 3D test cases, set the time interval for adding test cases, and set the upper limit for the number of test cases that can be run simultaneously. Increase the number of test cases that can be run at a fixed interval. When the GPU runs in a jammed or other overloaded state, record the number of test cases that can be run and stop the test.

[0080] Test the maximum number of video test cases played simultaneously: Run mainstream video test cases or customized video test cases, set the time interval for adding test cases, and the upper limit of the number of test cases to be run simultaneously. Increase the number of test cases to be run at a fixed interval. When the GPU is stuck or has excessive load, record the number of test cases to be run and stop the test.

[0081] For the specific test process, take the restart test as an example. During the specific test process of the performance test, first ensure that the physical connection between the GPU chip to be tested and the hardware platform is normal, and then set the test parameters in the console. Set the total number of restarts to 50 times, the interval between each restart is 60S, and the restart time is after the restart times are completed. Record the hardware circuit data and test time information during the test. During the restart process, read the GPU power-on to stable operation, and power off after stable operation for 60S. During the whole process, the GPU detection system feedbacks the changes in current, voltage, and temperature, and records the change trend, minimum value, and maximum value. After 50 restarts, there will be 50 GPU restart data. Normalize these 50 data, draw a regression curve, observe the changes in GPU current, voltage, temperature during these 50 restarts, the consistency of the change trend, minimum value, maximum value and other parameters, and analyze the fluctuation range.

[0082] For the second part, the data is processed graphically, combined with Figure 5 As shown, the specific process is as follows:

[0083] Read the excel1 file and draw relevant data graphs based on the test data fed back by the hardware platform during the routine performance test and the extreme performance test to show the changes in current, voltage, temperature, and power consumption of the GPU during the running of different test cases. The data graphs are divided into two categories, namely, single test graphical display and X / Y axis comparison graphical display of multiple test cases.

[0084] In this section, during the aforementioned routine performance test and extreme performance test, the console reads the current, voltage, temperature, power consumption and other parameters in the hardware circuit in real time through the communication protocol, associates the test data with the current software test function and specific use cases, and records the specific test cases and test results in the same excel file. The file is named with the hardware + test dimensions, and saved in the excel main file named GPU. During the performance test, the test data stored in the excel file is processed in the console, and the relevant data graphs are drawn according to the test data of each test case in the automatic test case during the routine performance test and the extreme performance test. The changes in current, voltage, temperature, power consumption, etc. of the GPU graphics card during the operation of different test cases. Graphical display is divided into two categories, single test graphical display and X / Y axis comparison graphical display of multiple test cases.

[0085] Here is an example of 2D performance testing:

[0086] At a certain time interval, the 2D performance test process of the GPU graphics card is performed multiple times. The current, voltage, temperature, power consumption and other single test graphical displays can be drawn to show the changes from power-on to the end of the test process in each test process, and the status of the hardware circuit during the GPU test process can be intuitively displayed. When the number of tests is sufficient, multiple test graphs of the same test process can be obtained to obtain the change trend of the current, voltage, temperature and power consumption of the GPU chip when running the 2D display, providing data support for subsequent GPU applications. When running multiple 2D test cases, the variables can be controlled to control that only the number of cases run during the test process changes. The current, voltage and temperature change trends of a single running case and multiple running cases are drawn in one graph to compare and analyze the impact of the parameters in the GPU peripheral hardware circuit during the 2D running process. Different test items can also be put together for comparison. Testers can freely call and process data to obtain the test comparison data they want, GPU chip operation and performance parameters.

[0087] For the third part, a GPU test database is established. The specific process is as follows:

[0088] During extreme performance testing, data is collected and analyzed for abnormal conditions that occur during testing (such as screen distortion, freezes, black screens, and other unstable performance conditions). Fault cases are established during the automatic performance testing process to provide preliminary debugging directions and locate problems for abnormal problems that may arise in subsequent GPU chip product applications.

[0089] In summary, first of all, the GPU automated testing method and system combined with software and hardware provided by the present invention has constructed an automated testing method for domestic GPU chips on multiple platforms, which can screen GPU chips and set parameters such as test cases, test items, and test time. The function and performance of GPU chips can be tested, and the performance of GPU chips can be investigated, which can be applied to the product warehousing and selection of GPU chips.

[0090] Secondly, the present invention reads the relevant current, voltage, and temperature parameters of the hardware platform during the software automation test, and jointly analyzes the data in the hardware platform and associates it with the performance displayed during the software test. The console processes the test data, graphically analyzes the data, collects and analyzes the data of abnormal situations that occur during the test, and establishes a test failure case library. The test failure case library uses a large amount of test data for comparative analysis. The comparative analysis results can preliminarily locate the abnormal problems that occur during the test, provide data support and theoretical basis for the subsequent practical application process of the GPU, improve the GPU detection efficiency, and ensure the subsequent normal application of the GPU.

[0091] Third, in the extreme performance test project, a large amount of test data is used for comparative analysis. For the functional abnormalities that may occur in the GPU during the test, such as screen distortion, freeze, black screen and other unstable performance conditions, the corresponding hardware circuit data is saved and analyzed in real time. A fault case database is established to provide preliminary debugging directions for abnormal problems that may occur in subsequent GPU chip product applications and locate the problems.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A GPU automated testing system combining software and hardware, characterized in that: The GPU automated testing system includes a hardware platform and a software platform. The software platform is used to set the voltage source output voltage value, power-on time and power-off time of the hardware platform, and obtain the current, voltage and temperature data fed back by the hardware platform, and associate the data with the automated test GPU performance, establish a GPU test fault case database, analyze the specific parameters and influencing factors that affect the GPU function and performance, locate the abnormal problems that occur during the test, and realize GPU automated testing; The hardware platform is used to provide GPU power supply, and control GPU power on and off and GPU power supply voltage value according to the settings of the software platform, and read the real-time current, voltage and temperature data near the GPU pins and feed them back to the software platform; The hardware platform includes a voltage source, a voltage acquisition module, a current acquisition module, and a temperature reading module. The voltage acquisition module, the current acquisition module, and the temperature reading module are respectively connected to the GPU to be tested, and the voltage source supplies power to the GPU; the software platform is implemented through a console, and the console is connected to the voltage source and the voltage acquisition module, the current acquisition module, and the temperature reading module.

2. The GPU automated testing system combining software and hardware as claimed in claim 1, characterized in that: The console is a module or chip with communication interaction, display, and fast computing capabilities, and includes: A voltage source regulating unit, used to realize digital control of the voltage source and perform real-time regulation of the voltage source; A calculation and analysis unit, used to store, calculate and analyze the voltage data output by the voltage acquisition module, the current data output by the current acquisition module, and the GPU shell temperature data output by the temperature reading module; The graphical unit is used to display and compare the output voltage value of the voltage source, the voltage data output by the voltage acquisition module, the current data output by the current acquisition module, and the temperature data output by the temperature reading module in real time, and draw corresponding curves to show the fluctuation range and change trend of the voltage, current, power consumption, and temperature when the GPU is working; The test unit is used to perform performance testing on the GPU. During the test, the output voltage value of the voltage source, as well as the GPU pin voltage value, current value, temperature value, and power consumption are monitored in real time, and the test results are associated with the monitoring data.

3. The GPU automated testing system combining software and hardware as claimed in claim 2, characterized in that: The voltage source includes an external reference power supply, a precision unit and a voltage superposition circuit. The voltage source provides a power supply voltage for the GPU. Multiple precision units are connected in series between the external reference power supply and the ground. The total resistance of each precision unit is fixed. Each precision unit is divided according to the resistance value. The precision unit controls the connected resistance value through a channel selector to change the connected power supply voltage value.

4. A GPU automated testing method combining software and hardware, characterized in that: The GPU automated testing method is implemented based on the GPU automated testing system according to any one of claims 1 to 3. The GPU automated testing method includes: Step S1, GPU chip performance test: using the test case library and example execution instruction scripts to automatically test GPU computing and video memory pressure; Step S2, data graphical processing: after the GPU chip performance test is completed, the test data is processed by using software tools, combined with the data fed back by the hardware platform, for graphical display and data graphical analysis; Step S3, establishing a GPU test database: collecting and analyzing data on abnormal situations that occur during the test, and establishing a test failure case library.

5. The GPU automated testing method combining software and hardware as claimed in claim 4, characterized in that: The specific process of the GPU chip performance test is as follows: S11. Design test items and build a complete set of test cases according to the localized GPU chip platform and localized GPU models, and deploy automated testing tools; S12, after selecting the test items and test cases for the software test of the current GPU chip, set the test parameters, including the test interval, the number of tests, and the test running time; S13. Generate test execution instructions according to the selected test items and test cases, as well as the set test parameters, and obtain the test case script file of the test instance from the specified directory; S14, executing the test case script file to perform automated testing, and displaying the current test status locally in real time. After the automated testing is completed, the test results are generated into a test report, and the test data is saved; At the same time, during the test, the current, voltage, temperature, and power consumption data fed back by the hardware platform are read in real time, and the changing trend, minimum value, and maximum value of the data are recorded.

6. The GPU automated testing method combining software and hardware as claimed in claim 5, characterized in that: The selected test cases include conventional performance tests and extreme performance tests. Conventional performance tests include restart test cases, 2D performance test cases, 3D test cases, pixel fill rate test cases, texture fill rate test cases and computing power test cases. Extreme performance tests include using a multi-process method to continuously increase the number of 2D test cases, 3D test cases and video playback, and at the same time determine whether the number of corresponding test case processes increases synchronously, to obtain the maximum number of 2D test cases, 3D test cases and video playback running simultaneously, and save the test cases and test results in excel files named after the current test dimension, and save them in the excel total file named after GPU.

7. The GPU automated testing method combining software and hardware as claimed in claim 6, characterized in that: The specific process of data graphical processing is as follows: Read the excel1 file and draw relevant data graphs based on the test data fed back by the hardware platform during the routine performance test and the extreme performance test to show the changes in current, voltage, temperature, and power consumption of the GPU during the running of different test cases. The data graphs are divided into two categories, namely, single test graphical display and X / Y axis comparison graphical display of multiple test cases.

8. The GPU automated testing method combining software and hardware as claimed in claim 7, characterized in that: The specific process of establishing a GPU test database is as follows: During the extreme performance test, data on abnormal situations that occur during the test is collected and analyzed, and a GPU test fault case library is established to provide preliminary debugging directions for abnormal problems that may arise in subsequent GPU chip product applications and to locate the problems.

Images