CN104268046A - Linux manual interaction NVIDIA GPU automatic testing method - Google Patents

Abstract

The invention discloses an NVIDIA GPU automatic testing method of manual interaction under Linux, which comprises the following concrete implementation processes: building a test platform; preparing a test environment; confirming the number of GPUs; installing a GPU driving tool; confirming GPU information; operating pressure test and city-springing monitoring; and finishing the test after confirming the result. Compared with the prior art, the Linux manual interaction NVIDIA GPU automatic testing method has the advantages that the NVIDIA GPU can achieve double-layer verification effect through the information captured by the system from hardware and manual interaction confirmation, the existence of unidentified GPU cards can be automatically detected and judged, the practicability is high, the stability of a server is effectively ensured, and the Linux manual interaction NVIDIA GPU automatic testing method is a very effective method for verifying the product quality.

Description

The NVIDIA GPU automated testing method of man-machine interactively under a kind of Linux

Technical field

The present invention relates to server accessory technical field of measurement and test, specifically the NVIDIA GPU automated testing method of man-machine interactively under a kind of practical, Linux.

Background technology

Along with the high speed development of server industries, in every field, increasing client adopts large batch of server to run the core application of oneself.Each server provider domestic at present, in order to meet the client of graphics calculations and large-scale PC cluster demand, releases the high-performance server of the stronger configuration NVIDIA GPU of many moneys computing power.

For the server system that NVIDIA GPU configures, majority carries out debugging configuration under system and pressure test based on manual mode, will strengthen the input of manpower and time like this, and artificial input complicated order easily produces mistake simultaneously.

Along with the continuous increase of server shipment amount, outside the stability ensureing product and reliability, the procedure of Product checking, robotization, high efficiency also become essential.

GPU supports T & L(TransformandLighting from hardware, polygon conversion and light source process), and T & L is a pith during 3D plays up, and also can be called " geometric manipulations ".GPU is no longer confined to 3D graphics process simultaneously, and the fact also proves in the calculating of the part such as floating-point operation, parallel computation, and even GPU can provide decades of times hundreds of times in the performance of CPU.

In the 3D graphics process of high-performance business and floating-point operation, parallel computation, GPU can produce a large amount of heat radiations, and it is particularly important that the heat radiation of GPU and capability bandwidth test become.

Based on this, now provide the NVIDIA GPU automated testing method of man-machine interactively under a kind of Linux, the method realizes by writing automatic test script, reduces man-machine interactively process as far as possible, realizes the object that pipelining detects, practical.

Summary of the invention

Technical assignment of the present invention is for above weak point, provides the NVIDIA GPU automated testing method of man-machine interactively under a kind of practical, Linux.

A NVIDIA GPU automated testing method for man-machine interactively under Linux, its specific implementation process is:

One, build test platform, NVIDIA GPU is installed to server;

Two, setup test environment;

Three, confirm that GPU is identified quantity, until all confirming to enter next step normally;

Four, GPU is installed drive and automatically load CUDA Install and configure;

Five, confirm and judge whether GPU information conforms to, and conforms to, and enters next step;

Six, operating pressure test, to the amendment of script argument option in test process, to realize carrying out the test item of different pressures value and different test durations;

Seven, check whether confirmation result meets GPU factory calibration;

Eight, test terminates.

Described test environment comprises installing operating system, selects disk partition and software package, and described operating system refers to Redhat linux system or Centos system.

In described step 3, GPU identifies that the judgement of quantity is completed by man-machine interactively, and this identifying comprises:

Judged by the quantity of system command to NVIDIA GPU;

Input the actual GPU quantity of order manually to compare, correctly then automatically carry out next step.

The detailed process of described step 4 is:

In/etc/rc.local file, write shielding firewall, selinux and deletion system carry video driver;

Auto-mounting GPU drives, and failure automatically generates log file for analysis, successful then loading CUDA Install and configure automatically.

In described step 5, the judgement man-machine interactively of GPU information completes, and its detailed process is:

Auto-mounting freeglut, cuda drive, failed then automatically generation log file for analysis;

By the environment configurations write/root/.bashrc file in PATH and the LD_LIBRARY_PATH path of CUDA;

Automatic display GPU information model, failed then point out check, successful then autoboot, and enter next step.

The NVIDIA GPU automated testing method of man-machine interactively under a kind of Linux of the present invention, has the following advantages:

Under a kind of Linux of this invention, the NVIDIA GPU automated testing method of man-machine interactively is by system command and automatically load CUDA TOOL software and detect GPU quantity and model, and in addition man-machine interactively confirms, improves the precision detected; Use automatic test script, by driving GPU, the Auto-mounting of environment configurations and pressure test runs, and greatly reduces the input of manpower and time, improves production capacity; Realize the procedure of NVIDIA GPU server product detection, robotization, high efficiency, practical, applied widely, be easy to promote.

Accompanying drawing explanation

Accompanying drawing 1 is realization flow figure of the present invention.

Embodiment

Below in conjunction with the drawings and specific embodiments, the invention will be further described.

The invention provides the NVIDIA GPU automated testing method of man-machine interactively under a kind of Linux, as shown in Figure 1, its specific implementation process is:

One, build test platform, NVIDIA GPU is installed to server;

Two, setup test environment;

Three, confirm that GPU is identified quantity, until all confirming to enter next step normally;

Four, GPU is installed drive and automatically load CUDA Install and configure;

Five, confirm and judge whether GPU information conforms to, and conforms to, and enters next step;

Six, operating pressure test, to the amendment of script argument option in test process, to realize carrying out the test item of different pressures value and different test durations;

Seven, check whether confirmation result meets GPU factory calibration;

Eight, test terminates.

Described test environment comprises installing operating system, selects disk partition and software package, and described operating system refers to Redhat linux system or Centos system.

In described step 3, GPU identifies that the judgement of quantity is completed by man-machine interactively, and this identifying comprises:

Judged by the quantity of system command to NVIDIA GPU;

Input the actual GPU quantity of order manually to compare, correctly then automatically carry out next step.

Part content for script is as follows:

a=`lspci |grep -i "3D controller: nVidia" |wc -l`

echo -n "Please Enter Physical GPU Number [1/2]:"

read Num

case ${Num} in

1)

if [ ! $a == 1 ]; then

echo -e "\033[41;37m GPU missing \033[0m"

else

./gpu.sh

Fi。

The detailed process of described step 4 is:

In/etc/rc.local file, write shielding firewall, selinux and deletion system carry video driver;

Auto-mounting GPU drives, and failure automatically generates log file for analysis, successful then loading CUDA Install and configure automatically.

Part content for script is as follows:

setenforce 0

echo "setenforce 0" >>/etc/rc.local

echo "iptables -F" >>/etc/rc.local

echo "rmmod nouveau" >>/etc/rc.local

sh NVIDIA-Linux-x86_64-319.60.run -a -s --no-x-check

if [$!= 0 ]; then

echo -e "\033[31m \033[05m GPU driver install FAIL\033[0m"

echo "GPU driver FAIL" >> /root/gpu_log.txt

else

sh /home/cuda.sh。

In described step 5, the judgement man-machine interactively of GPU information completes, and its detailed process is:

Auto-mounting freeglut, cuda drive, failed then automatically generation log file for analysis;

By the environment configurations write/root/.bashrc file in PATH and the LD_LIBRARY_PATH path of CUDA;

Automatic display GPU information model, failed then point out check, successful then autoboot, and enter next step.

Part content for script is as follows:

sh cuda_5.5.22_linux_64.run -silent -toolkit -samples

if [$!= 0 ]; then

echo -e "\033[31m \033[05m CUDA install fail\033[0m"

echo "cuda install fail" >>/root/gpu_log.txt

fi

echo "export PATH=/usr/local/cuda-5.5/bin:$PATH" >>/root/.bashrc

echo "export LD_LIBRARY_PATH=/usr/local/cuda-5.5/lib64:$LD_LIBRARY_PATH" >>/root/.bashrc

source /root/.bashrc

cd /usr/local/cuda-5.5/samples

make

/usr/local/cuda-5.5/samples/bin/x86_64/linux/release/deviceQuery |tee /home/gpu.txt

cat/home/gpu.txt|awk"/\"TeslaK20m\"/{print}/globalmemory/{print}/CUDAores/{print}/Clock rate:/{print}/deviceQuery,/{print}/Result/{print}"

echo -n "Please Check GPU infomation [OK/FAIL]:"

read Status

case ${Status} in

OK | ok)

Reboot。

The NVIDIA GPU automated testing method of man-machine interactively of the present invention, be suitable for the operating system such as the red cap of x86 framework and Centos, the information that the method is grabbed from hardware by system and man-machine interactively confirm, double-deck verification the verifying results can be reached to NVIDIA GPU, and automatically can detect and determine whether Unidentified GPU card.

This method of testing is simple to operate, greatly reduce the time (installing together with system, foreshortening within 1 hour from being greater than 2 hours) of producing the operation of line manual detection and intervening, practical, effectively ensure that the stability of server, is the very effective method of checking product quality.

This method of testing Successful utilization in the debugging of producing line configuration NVIDIA GPU server product and pressure burn-in test.

Above-mentioned embodiment is only concrete case of the present invention; scope of patent protection of the present invention includes but not limited to above-mentioned embodiment; under any a kind of Linux according to the invention the NVIDIA GPU automated testing method of man-machine interactively claims and any person of an ordinary skill in the technical field to its suitable change done or replacement, all should fall into scope of patent protection of the present invention.

Claims (5)

1. the NVIDIA GPU automated testing method of man-machine interactively under Linux, is characterized in that its specific implementation process is:

One, build test platform, NVIDIA GPU is installed to server;

Two, setup test environment;

Three, confirm that GPU is identified quantity, until all confirming to enter next step normally;

Four, GPU is installed drive and automatically load CUDA Install and configure;

Five, confirm and judge whether GPU information conforms to, and conforms to, and enters next step;

Six, operating pressure test, to the amendment of script argument option in test process, to realize carrying out the test item of different pressures value and different test durations;

Seven, check whether confirmation result meets GPU factory calibration;

Eight, test terminates.

2. the NVIDIA GPU automated testing method of man-machine interactively under a kind of Linux according to claim 1, it is characterized in that: described test environment comprises installing operating system, selects disk partition and software package, and described operating system refers to Redhat linux system or Centos system.

3. the NVIDIA GPU automated testing method of man-machine interactively under a kind of Linux according to claim 2, is characterized in that: in described step 3, GPU identifies that the judgement of quantity is completed by man-machine interactively, and this identifying comprises:

Judged by the quantity of system command to NVIDIA GPU;

Input the actual GPU quantity of order manually to compare, correctly then automatically carry out next step.

4. the NVIDIA GPU automated testing method of man-machine interactively under a kind of Linux according to claim 2, is characterized in that: the detailed process of described step 4 is:

In/etc/rc.local file, write shielding firewall, selinux and deletion system carry video driver;

Auto-mounting GPU drives, and failure automatically generates log file for analysis, successful then loading CUDA Install and configure automatically.

5. the NVIDIA GPU automated testing method of man-machine interactively under a kind of Linux according to claim 2, is characterized in that: in described step 5, the judgement man-machine interactively of GPU information completes, and its detailed process is:

Auto-mounting freeglut, cuda drive, failed then automatically generation log file for analysis;

By the environment configurations write/root/.bashrc file in PATH and the LD_LIBRARY_PATH path of CUDA;

Automatic display GPU information model, failed then point out check, successful then autoboot, and enter next step.