ShoufaChen / gpu-burn

Multi-GPU CUDA stress test

Building with custom CUDA_PATH and Compute Capability

cd gpu-burn
make COMPUTE=<compute capability value> CUDAPATH=/usr/local/cuda-<version>

Check compute capability for nvidia gpu: https://en.wikipedia.org/wiki/CUDA

GPU	CC
Tesla V100,	7.0
A100 80GB, A100 40GB	8.0

GPU Stress Test

./gpu_burn -d <num_seconds>

gpu-burn

Multi-GPU CUDA stress test http://wili.cc/blog/gpu-burn.html

Easy docker build and run

git clone https://github.com/wilicc/gpu-burn
cd gpu-burn
docker build -t gpu_burn .
docker run --rm --gpus all gpu_burn

Building

To build GPU Burn:

make

To remove artifacts built by GPU Burn:

make clean

GPU Burn builds with a default Compute Capability of 5.0. To override this with a different value:

make COMPUTE=<compute capability value>

CFLAGS can be added when invoking make to add to the default list of compiler flags:

make CFLAGS=-Wall

LDFLAGS can be added when invoking make to add to the default list of linker flags:

make LDFLAGS=-lmylib

NVCCFLAGS can be added when invoking make to add to the default list of nvcc flags:

make NVCCFLAGS=-ccbin <path to host compiler>

CUDAPATH can be added to point to a non standard install or specific version of the cuda toolkit (default is /usr/local/cuda):

make CUDAPATH=/usr/local/cuda-<version>

CCPATH can be specified to point to a specific gcc (default is /usr/bin):

make CCPATH=/usr/local/bin

Usage

GPU Burn
Usage: gpu_burn [OPTIONS] [TIME]

-d	Use doubles
-tc	Use Tensor cores
-h	Show this help message

Example:
gpu_burn -d 3600

Example

cd gpu-burn
make COMPUTE=7.0 CUDAPATH=/usr/local/cuda-11.0

while :
do
  ./gpu_burn -d 180
  sleep 1800
done

sleep 10h