Method and device for automatically verifying NPU drive stability and performance

Abstract

The invention relates to a method and device for automatically verifying NPU drive stability and performance. The method comprises the steps that at least one NPU card is installed in a to-be-tested server; the upper computer starts the to-be-tested server by controlling the source controller, and the upper computer controls the to-be-tested server to install a system and a dependency package; executing an Int8 computing power test; executing a pressure test; the upper computer controls the power supply controller to execute power-off and power-on operations on the to-be-tested server; the upper computer controls the server to be tested to install the driver and firmware of the next version; and the upper computer analyzes and counts data. According to the method and the device for automatically verifying the stability and the performance of the NPU driver, different versions of drivers and firmware can be automatically switched. Through the control of the power supply controller, the pressure test is realized, and the compatibility and stability of the drive and the system are verified. The computing power of the NPU card is tested, the actual computing power is obtained, and meanwhile the functions of testing and calculating floating power consumption are achieved.

Description

Technical Field

[0001] The present invention relates to the field of graphics card testing, and in particular to a method and device for automatically verifying NPU driver stability and performance. Background Art

[0002] NVIDIA's GPU is developing rapidly in the field of AI, and the development of AI cards is also progressing rapidly in China. Since the AI ​​field mainly does matrix operations, Huawei has designed an AI chip specifically for matrix operations - the Ascend series (NPU). NPU cards have powerful performance in training and reasoning, but due to the large number of product models and rapid development, the adaptability of the system, driver and firmware to the card has become increasingly important, and to a large extent determines the stability and performance of the AI ​​card. When the NPU card is manufactured, the hardware structure is fixed, but the updated driver and firmware will be updated all the time. Due to the large number of card models and fast version iteration, the system and version specified by the customer are automatically implemented and verified, which has solved the process of manual verification. The verification process of different versions of manual testing drivers and firmware is very long, and the uninstallation and installation process is also long and cumbersome. At the same time, there is a lack of analysis process for actual computing power and power consumption. This automated testing process, while covering the traditional testing process, optimizes the process and time, and optimizes computing power and power consumption. Summary of the invention

[0003] The present invention provides a method and a testing device for automatically verifying NPU driving stability and performance, aiming to solve at least one of the technical problems existing in the prior art.

[0004] The technical solution of the present invention is a method for automatically verifying the stability and performance of an NPU driver. The method for automatically verifying the stability and performance of an NPU driver is run on a device for automatically verifying the stability and performance of an NPU driver. The device for automatically verifying the stability and performance of an NPU driver includes a host computer, a power controller, and a server to be tested. At least one NPU card is installed in the server to be tested. The method for automatically verifying the stability and performance of an NPU driver includes the following steps: S100, installing at least one NPU card into the server to be tested, connecting the host computer and the power controller to the server to be tested respectively, and connecting the power controller to the host computer; S200, the host computer starts the server to be tested by controlling the power controller, and the host computer controls the server to be tested to install a system and dependent packages; S300, the host computer installs the driver and firmware on the server to be tested; S400, the host computer performs an Int8 computing power test on the server to be tested after the driver and firmware are installed; S500, the host computer performs a stress test on the server to be tested; S600, the host computer controls the power controller to perform power-off and power-on operations on the server under test, and the server under test after restarting repeats steps S400 and S500, and then repeats step S600 for a preset number of times and then jumps to step S700; S700, the host computer controls the server to be tested to install the next version of the driver and firmware; S800: The host computer analyzes and collects statistical data.

[0005] Further, the step S200 includes: S210, the host computer controls the server to be tested to install an operating system, and the operating system at least includes openeuler 22.03sp3; S220, the host computer controls the server to be tested to install dependent libraries and Python compilation libraries.

[0006] Further, in step S220, The dependent libraries at least include gcc, gcc-c++, make, cmake, unzip, zlib-devel, libffi-devel, openssl-devel, pciutils, net-tools, sqlite-devel, lapack-devel, gcc-gfortran and python3-devel; Install Python compiled libraries including at least attrs, numpy, decorator, sympy, cffi, pyyaml, pathlib2, psutil, protobuf, scipy, requests, and absl-py.

[0007] Further, the step S300 includes: S310, use the scp command to copy and install the driver and firmware through two IP addresses; S320, after installing the driver and firmware, install the toolkit and toolbox; S330. The host computer verifies whether the driver and firmware are successfully installed through the npu-smi info command.

[0008] Further, in step S400, the peak computing power of the NPU card is obtained by using the ascend-dmi -f -t int8 command.

[0009] Further, the step S500 includes: S510, perform a stress test for a preset time by using the ascend-dmi -p -dur xx -it X -pm refresh command; S520. Execute stdbuf -oL npu-smi info watch -i 0 -c 0 >v1.log 2>&1 to count the output computing power and calculate the average value and floating range of the power consumption.

[0010] Further, the step S700 includes: S710. Execute Ascend-hdk-xxx.run --uninstall to first uninstall the firmware currently installed on the server to be tested, and then uninstall the driver currently installed on the server to be tested. S720, repeating step S300, the host computer installing a new driver and a new firmware on the server to be tested; S730, repeating steps S400 to S600, the host computer performs an Int8 computing power test, a stress test, and a preset number of power-off and power-on operations on the new driver and the new firmware newly installed on the server to be tested.

[0011] Further, the step S800 includes: S810, determining the power consumption floating error through a stress test; S820, determining average power consumption through a stress test; S830, calculate the maximum computing power through the Int8 computing power test; S840. Calculate the maximum energy consumption ratio based on the preset computing power weight and power consumption weight.

[0012] Furthermore, the present invention also proposes a device for automatically verifying the stability and performance of an NPU driver, wherein the device for automatically verifying the stability and performance of an NPU driver is used to implement the method for automatically verifying the stability and performance of an NPU driver, and the device for automatically verifying the stability and performance of an NPU driver comprises: Host computer; A power supply controller, the host computer is electrically connected to the power supply controller; The server to be tested, the host computer and the power controller are respectively connected to the server to be tested, the server to be tested is an X86 model server or an ARM model server, and the server to be tested includes at least one central processing unit, one memory bar, one PCIE slot and one hard disk; At least one NPU card is installed in the server to be tested, and the NPU card is connected to the server to be tested via a PCIE interface.

[0013] Furthermore, the present invention also proposes a computer-readable storage medium having program instructions stored thereon, and when the program instructions are executed by a processor, the method of automatically verifying the stability and performance of the NPU driver is implemented.

[0014] The beneficial effects of the present invention are: The method and device for automatically verifying the stability and performance of the NPU driver can automatically switch between different versions of drivers and firmware, including the installation and uninstallation of the driver and firmware of the server to be tested, and realize stress testing by using the power controller to control and verify the compatibility and stability of the driver and system. The test of the computing power of the NPU card obtains the actual computing power, and has the function of testing and calculating the floating power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of a method for automatically verifying the stability and performance of an NPU driver.

[0016] Figure 2 This is a connection control structure diagram of a device that automatically verifies the stability and performance of the NPU driver.

[0017] Figure 3 It is a schematic diagram of the structure of a server to be tested in a device for automatically verifying the stability and performance of an NPU driver.

[0018] Figure 4 It is a schematic diagram of different versions of drivers and firmware, as well as tool sets and toolkit files in a method for automatically verifying NPU driver stability and performance.

[0019] Figure 5 This is a schematic diagram of a driver and firmware folder of one version in the method of automatically verifying the stability and performance of the NPU driver, which contains a specific version of the driver file and firmware file.

[0020] Figure 6 It is a schematic diagram of an implementation process of a method for automatically verifying the stability and performance of an NPU driver.

[0021] Figure 7 This is a schematic diagram of verifying whether the driver and firmware are successfully installed in a method for automatically verifying the stability and performance of the NPU driver.

[0022] Figure 8 It is a schematic diagram of calculating the peak computing power in the method of automatically verifying the stability and performance of the NPU driver.

[0023] Fig. 9 This is a schematic diagram of the difference in peak computing power results caused by different versions of drivers in the method of automatically verifying the stability and performance of the NPU driver.

[0024] Fig.10 It is a schematic diagram of the utilization rate of the NPU in the method of automatically verifying the stability and performance of the NPU driver.

[0025] Fig.11 It is a schematic diagram of average power consumption and floating error of NPU in the method of automatically verifying NPU driving stability and performance. DETAILED DESCRIPTION

[0026] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention, so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0027] It should be noted that, unless otherwise specified, when a feature is referred to as being "fixed" or "connected" to another feature, it may be directly fixed or connected to the other feature, or it may be indirectly fixed or connected to the other feature. In addition, the descriptions of up, down, left, right, top, bottom, etc. used in the present invention are only relative to the relative positional relationship of the components of the present invention in the drawings.

[0028] In addition, unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art. The terms used in this specification are only for describing specific embodiments, not for limiting the present invention. The term "and / or" used herein includes any combination of one or more related listed items.

[0029] It should be understood that, although the terms first, second, third, etc. may be used to describe various elements in the present disclosure, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from each other. For example, without departing from the scope of the present disclosure, the first element may also be referred to as the second element, and similarly, the second element may also be referred to as the first element.

[0030] Reference Figures 1 to 11 In some embodiments, the technical solution of the present invention is a method for automatically verifying the stability and performance of an NPU driver. The method for automatically verifying the stability and performance of an NPU driver is run on a device for automatically verifying the stability and performance of an NPU driver. The device for automatically verifying the stability and performance of an NPU driver includes a host computer, a power controller, and a server to be tested. At least one NPU card is installed in the server to be tested. Figure 1 and Figure 6 The method for automatically verifying the stability and performance of the NPU driver comprises the following steps: S100, installing at least one NPU card into the server to be tested, connecting the host computer and the power controller to the server to be tested respectively, and connecting the power controller to the host computer; S200, the host computer starts the server to be tested by controlling the power controller, and the host computer controls the server to be tested to install a system and dependent packages; S300, the host computer installs the driver and firmware on the server to be tested; S400, the host computer performs an Int8 computing power test on the server to be tested after the driver and firmware are installed; S500, the host computer performs a stress test on the server to be tested; S600, the host computer controls the power controller to perform power-off and power-on operations on the server under test, and the server under test after restarting repeats steps S400 and S500, and then repeats step S600 for a preset number of times and then jumps to step S700; S700, the host computer controls the server to be tested to install the next version of the driver and firmware; S800: The host computer analyzes and collects statistical data.

[0031] The beneficial effects of the present invention are: The method and device for automatically verifying the stability and performance of the NPU driver can automatically switch between different versions of drivers and firmware, including the installation and uninstallation of the driver and firmware of the server to be tested, and realize stress testing by using the power controller to control and verify the compatibility and stability of the driver and system. The test of the computing power of the NPU card obtains the actual computing power, and has the function of testing and calculating the floating power consumption.

[0032] The method and device for automatically verifying the stability and performance of the NPU driver replaces the manual verification test operation and performs automated operations on the driver and firmware stability of the AI ​​card. The power consumption and power consumption stability are considered while testing the computing power. When multiple versions and driver firmware are iterated, the most suitable driver firmware version can be selected.

[0033] Specifically, in step S100, the hardware is assembled to prepare a server to be tested (which can be an X86 model server - Powerleader self-developed server, or an ARM model server Kunpeng server) including major configurations such as CPU, memory, hard disk, and one or more NPU cards (depending on the customer's actual assembly requirements).

[0034] Further, the step S200 includes: S210, the host computer controls the server to be tested to install an operating system, and the operating system at least includes openeuler 22.03sp3; S220, the host computer controls the server to be tested to install dependent libraries and Python compilation libraries.

[0035] Specifically, in step S200, the system and dependent packages are installed, the operating system required by the customer, such as openeuler 22.03sp3, is installed, and then the system dependent libraries and the support of the Python compilation library are installed.

[0036] Including installing system dependent libraries: yum install -y gcc gcc-c++ make cmake unzip zlib-devel libffi-devel openssl-devel pciutils net-tools sqlite-devel lapack-develgcc-gfortran python3-devel It also includes installing python3 dependent libraries: pip3 install attrs pip3 install numpy pip3 install decorator pip3 install sympy pip3 install cffi pip3 install pyyaml pip3 install pathlib2 pip3 install psutil pip3 install protobuf pip3 install scipy pip3 install requests pip3 install absl-py Further, in step S220, The dependent libraries at least include gcc, gcc-c++, make, cmake, unzip, zlib-devel, libffi-devel, openssl-devel, pciutils, net-tools, sqlite-devel, lapack-devel, gcc-gfortran and python3-devel; Install Python compiled libraries including at least attrs, numpy, decorator, sympy, cffi, pyyaml, pathlib2, psutil, protobuf, scipy, requests, and absl-py.

[0037] Specifically, installing dependent libraries includes at least the following commands: yum install -y gcc gcc-c++ make cmake unzip zlib-devel libffi-developenssl-devel pciutils net-tools sqlite-devel lapack-devel gcc-gfortranpython3-devel.

[0038] Further, the step S300 includes: S310, use the scp command to copy and install the driver and firmware through two IP addresses; S320, after installing the driver and firmware, install the toolkit and toolbox; S330. The host computer verifies whether the driver and firmware are successfully installed through the npu-smi info command.

[0039] Specifically, install the driver and firmware: read the directory from the firmware source address, select the first version, automatically install the driver and firmware (here is 6.0.1.alpha001), and then install the toolkit and toolbox. The specific implementation is to use the scp command to copy through two IP addresses, and then execute Ascend-hdk-xxx.run --full, Ascend-hdk-xxx.run –install.

[0040] Specifically, the dependency packages of the Ascend environment are described as follows. In addition to the operating system, the Ascend environment has 4 dependency packages. The method verification test for automatically verifying the stability and performance of the Ascend NPU driver is to switch the driver and firmware versions, and the tool set and toolkit versions remain unchanged.

[0041] Reference Figure 4 , which contains multiple different versions of drivers and firmware. Each round of testing installs the drivers and firmware in a folder in sequence. The two run files are the toolkit and toolbox respectively. These two run files do not need to be replaced and only need to be installed once.

[0042] Reference Figure 5 In a specific version of the driver and firmware folder, there are two files, namely the driver file and the firmware file.

[0043] Specifically, the included file types are as follows: 1) Driver (multiple versions): Ascend-hdk-xxx-npu-dirver_xxxx.run 2) Firmware (multiple versions): Ascend-hdk-xxx-npu-firmware_xxxx.run 3) Toolkit (version unchanged): Ascend-cann-toolkit_xxx.run 4) Toolbox (version unchanged): Ascend-mindx-toolbox_xxx.run Specifically, refer to Figure 7 In step S330, npu-smi info is used to verify whether the driver and firmware are successfully installed. When the driver and firmware are successfully installed, the command line will have default npu-smi parameters, and the hardware information can be viewed through npu-smi --help.

[0044] Further, in step S400, the peak computing power of the Ascend NPU card is obtained through the ascend-dmi -f -t int8 command.

[0045] Specifically, refer to Figure 8 , execute the ascend-dmi -f -t int8 command, which takes about 1.6 seconds and is recorded.

[0046] Reference Fig. 9 After installing different versions of drivers and firmware, different computing power test results were obtained through the Int8 computing power test.

[0047] Further, refer to Fig.10 and Fig.11 , the step S500 comprises: S510, perform a stress test for a preset time by using the ascend-dmi -p -dur xx -it X -pm refresh command; S520. Execute stdbuf -oL npu-smi info watch -i 0 -c 0 >v1.log 2>&1 to count the output computing power and calculate the average value and floating range of the power consumption.

[0048] Specifically, in step S510, referring to Fig.10 Table 1 shows the results of a 5-minute stress test, which shows that the NPU utilization rate has reached 100%.

[0049] Table 1, Among them, NpuID refers to the serial number of the card, ChipID refers to the serial number of the chip on the card, Pwr(W) refers to the power, Temp refers to the temperature of the card, AI core refers to the load of the AI ​​chip, AI CPU refers to the load of the logic chip on the card, Ctrl cpu refers to the load of the control cpu, Memory refers to the occupancy rate of the video memory, and Memory BW refers to the proportion of the video memory bandwidth.

[0050] In step S520, the average power consumption, maximum power consumption, minimum power consumption and floating error are obtained by executing the following code: fh = open('v1.log') l = [] def is_number(s): try: float(s) return True except ValueError: pass try: import unicodedata unicodedata.numeric(s) return True except (TypeError, ValueError): pass return False for index ,lx in enumerate(fh): if index >= 20: power = lx.rstrip().split()[2] if is_number(power): l.append(float(power)) fh.close() print("The average power consumption is:" + str(sum(l) / len(l))+ "w") print("Maximum power consumption:" + str(max(l))+ "w") print("Minimum power consumption:" + str(min(l))+ "w") diff_max = (max(l) - min(l)) / ((max(l) + min(l) / 2)) print("Floating error: " + str(diff_max)) In step S600, the preset number of times of re-execution of step S600 is set to 5 or 6. Two functions are mainly verified, the first is the stability verification of the drive, and the second is the repeated verification of the test data to ensure the accuracy of the data.

[0051] Further, the step S700 includes: referring to Fig.11 , the average value and floating range of power consumption are automatically analyzed through Python scripts.

[0052] S710, by executing Ascend-hdk-xxx.run –uninstall, first uninstall the firmware currently installed on the server to be tested, and then uninstall the driver currently installed on the server to be tested; S720, repeating step S300, the host computer installing a new driver and a new firmware on the server to be tested; S730, repeating steps S400 to S600, the host computer performs an Int8 computing power test, a stress test, and a preset number of power-off and power-on operations on the new driver and the new firmware newly installed on the server to be tested.

[0053] Further, the step S800 includes: S810, determining the power consumption floating error through a stress test; S820, determining average power consumption through a stress test; S830, calculate the maximum computing power through the Int8 computing power test; S840. Calculate the maximum energy consumption ratio based on the preset computing power weight and power consumption weight.

[0054] Specifically, in step S840, the energy efficiency ratio is calculated based on the computing power weight parameter W and the power consumption weight parameter Y. In a specific embodiment, the energy efficiency ratio = (computing power * W) / (power consumption * Y).

[0055] Furthermore, the present invention also proposes a device for automatically verifying the stability and performance of an NPU driver, wherein the device for automatically verifying the stability and performance of an NPU driver is used to implement the method for automatically verifying the stability and performance of an NPU driver, and the device for automatically verifying the stability and performance of an NPU driver comprises: Host computer; A power supply controller, the host computer is electrically connected to the power supply controller; The server to be tested, the host computer and the power controller are respectively connected to the server to be tested, the server to be tested is an X86 model server or an ARM model server, and the server to be tested includes at least one central processing unit, one memory bar, one PCIE slot and one hard disk; At least one NPU card is installed in the server to be tested, and the NPU card is connected to the server to be tested via a PCIE interface.

[0056] Specifically, refer to Figure 2 First of all, when testing the NPU card, you need to assemble the hardware structure of the NPU card. The Ascend NPU card is generally connected via PCIE. Therefore, when installing, you can directly insert the NPU card into the PCIE slot of the server to be tested.

[0057] Reference Figure 3 The server to be tested includes at least a central processing unit, a memory bar, a PCIE slot and a hard disk.

[0058] Reference Figure 3 The communication relationship between CPU, NPU, memory and hardware is supplemented as follows: First, the calculated data is stored in the hard disk, and the program executes to read the data into the memory for temporary storage. Then the data is copied from the memory to the video memory of the NPU card, and the AI ​​core of the NPU performs calculations on the data again. When the calculation results are obtained, the video memory of the NPU card is copied back to the system memory for subsequent data processing (data storage, non-convolution calculation processing of the central processing unit CPU).

[0059] Furthermore, the present invention also proposes a computer-readable storage medium having program instructions stored thereon, and when the program instructions are executed by a processor, the method of automatically verifying the stability and performance of the NPU driver is implemented.

[0060] In a specific embodiment, the power controller is a power controller of Toprise Technology, model XC3006, power 2500W. Some Chinese annotations are as follows: NPU: Huawei's Ascend card.

[0061] ascend-dmi, npu-smi: Huawei Ascend software commands.

[0062] stdbuf: is a Linux system command.

[0063] Furthermore, the present invention also provides a computer-readable storage medium on which program instructions are stored. When the program instructions are executed by a processor, the USB interface power automatic testing method is implemented.

[0064] The above is only a preferred embodiment of the present invention. The present invention is not limited to the above implementation. As long as the technical effect of the present invention is achieved by the same means, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the scope of protection of the present disclosure. All should belong to the protection scope of the present invention. Within the protection scope of the present invention, its technical scheme and / or implementation method can have various modifications and changes.

Claims

1. A method for automatically verifying the stability and performance of an NPU driver, wherein the method is run on a device for automatically verifying the stability and performance of an NPU driver, wherein the device comprises a host computer, a power controller and a server to be tested, wherein at least one NPU card is installed in the server to be tested, and wherein: The method for automatically verifying the stability and performance of the NPU driver comprises the following steps: S100, installing at least one NPU card into the server to be tested, connecting the host computer and the power controller to the server to be tested respectively, and connecting the power controller to the host computer; S200, the host computer starts the server to be tested by controlling the power controller, and the host computer controls the server to be tested to install a system and dependent packages; S300, the host computer installs the driver and firmware on the server to be tested; S400, the host computer performs an Int8 computing power test on the server to be tested after the driver and firmware are installed; S500, the host computer performs a stress test on the server to be tested; S600, the host computer controls the power controller to perform power-off and power-on operations on the server to be tested, and the server to be tested after restarting repeats steps S400 and S500, and then repeats step S600 for a preset number of times and then jumps to step S700; S700, the host computer controls the server to be tested to install the next version of the driver and firmware; S800: The host computer analyzes and collects statistical data.

2. The method for automatically verifying NPU driver stability and performance according to claim 1, characterized in that: The step S200 includes: S210, the host computer controls the server to be tested to install an operating system, and the operating system at least includes openeuler 22.03sp3; S220, the host computer controls the server to be tested to install dependent libraries and Python compilation libraries.

3. The method for automatically verifying NPU driving stability and performance according to claim 2, characterized in that: In the step S220, The dependent libraries at least include gcc, gcc-c++, make, cmake, unzip, zlib-devel, libffi-devel, openssl-devel, pciutils, net-tools, sqlite-devel, lapack-devel, gcc-gfortran and python3-devel; Install Python compiled libraries including at least attrs, numpy, decorator, sympy, cffi, pyyaml, pathlib2, psutil, protobuf, scipy, requests, and absl-py.

4. The method for automatically verifying NPU driver stability and performance according to claim 1, characterized in that: The step S300 includes: S310, use the scp command to copy and install the driver and firmware through two IP addresses; S320, after installing the driver and firmware, install the toolkit and toolbox; S330. The host computer verifies whether the driver and firmware are successfully installed through the npu-smi info command.

5. The method for automatically verifying NPU driver stability and performance according to claim 1, characterized in that: In step S400, the peak computing power of the NPU card is obtained by using the ascend-dmi -f -t int8 command.

6. The method for automatically verifying NPU driver stability and performance according to claim 1, characterized in that: The step S500 includes: S510, perform a stress test for a preset time by using the ascend-dmi -p -dur xx -it X -pm refresh command; S520. Execute stdbuf -oL npu-smi info watch -i 0 -c 0 >v1.log 2>&1 to count the output computing power and calculate the average value and floating range of the power consumption.

7. The method for automatically verifying NPU driver stability and performance according to claim 1, characterized in that: The step S700 includes: S710, by executing Ascend-hdk-xxx.run –uninstall, first uninstall the firmware currently installed on the server to be tested, and then uninstall the driver currently installed on the server to be tested; S720, repeating step S300, the host computer installing a new driver and a new firmware on the server to be tested; S730, repeating steps S400 to S600, the host computer performs an Int8 computing power test, a stress test, and a preset number of power-off and power-on operations on the new driver and the new firmware newly installed on the server to be tested.

8. The method for automatically verifying NPU driver stability and performance according to claim 1, characterized in that: The step S800 includes: S810, determining the power consumption floating error through a stress test; S820, determining average power consumption through a stress test; S830, calculate the maximum computing power through the Int8 computing power test; S840. Calculate the maximum energy consumption ratio based on the preset computing power weight and power consumption weight.

9. A device for automatically verifying the stability and performance of NPU drivers, characterized in that: The device for automatically verifying the stability and performance of the NPU driver is used to implement the method for automatically verifying the stability and performance of the NPU driver as claimed in any one of claims 1 to 8, and the device for automatically verifying the stability and performance of the NPU driver includes: Host computer; A power supply controller, the host computer is electrically connected to the power supply controller; The server to be tested, the host computer and the power controller are respectively connected to the server to be tested, the server to be tested is an X86 model server or an ARM model server, and the server to be tested includes at least one central processing unit, one memory bar, one PCIE slot and one hard disk; At least one NPU card is installed in the server to be tested, and the NPU card is connected to the server to be tested via a PCIE interface.

10. A computer-readable storage medium having program instructions stored thereon, wherein the program instructions implement the method according to any one of claims 1 to 8 when executed by a processor.

Images