• Changed!!! Supportted Nvidia JETSON GPU temperature information polling.

This tool thoroughly exercises the math units on NVIDIA GPUs. It is capable of stressing both the single-precision (32-bit) and double-precision (64-bit) math units of any GPU with CUDA support. Generally-speaking, this means any NVIDIA Tesla, Quadro or GeForce GPU released after 2010.

During startup, all GPUs in the system are discovered and tests are run on each. As the tool runs, the status of each GPU is output (including number of passes, number of errors encountered, and current GPU temperatures).

This tool must be compiled using GCC and NVIDIA CUDA. On most systems, all that will be necessary is to run make. The Makefile does its best to locate all the necessary dependencies. Once compiled, you can execute a test run with: ./gpu_burn

An RPM can be created by running the following from the base directory:

VERSION=0.4.6
fpm -t rpm -s dir --prefix=/usr                                             \
    --name gpu-burn -v ${VERSION}                                           \
    --vendor Microway --license GPLv3                                       \
    --url https://github.com/Microway/gpu-burn                              \
    --description 'Utility for stress-testing NVIDIA GPU accelerators'      \
     gpu_burn=sbin/ gpu_burn.cuda_kernel=libexec/

A DEB can be created by replacing the -t rpm above with -t deb.

By default, this utility runs for only a short time. The user must specify how many seconds the GPU stress test should be run.

To run a single-precision test for one hour, run:

gpu_burn $(( 60 * 60 ))

To run a double-precision test for one hour, run:

gpu_burn -d $(( 60 * 60 ))

If you would like to purchase professional support for gpu-burn, or to fund development of a new feature, please visit: https://www.microway.com/contact/

This project is based upon the work of Ville Timonen. This version of the tool is maintained by Microway, Inc.