Multi-GPU CUDA stress test (fork of wilicc/gpu-burn)
- adds support for FP16 (with and without tensor cores)
- adds support for FP16 multiply FP32 accumulate (requires tensor cores)
- change batch size for optimal FLOP/s
- more accurately calculate FLOP/s
- report TFLOP/s instead of GFLOP/s
- implement FP16 data comparison (check for errors)
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
CUDA_VERSION and IMAGE_DISTRO can be used to override the base
images used when building the Docker image target, while IMAGE_NAME
can be set to change the resulting image tag:
make IMAGE_NAME=myregistry.private.com/gpu-burn CUDA_VERSION=12.0.1 IMAGE_DISTRO=ubuntu22.04 image
GPU Burn
Usage: gpu_burn [OPTIONS] [TIME]
-m X Use X MB of memory
-m N% Use N% of the available GPU memory
-fp64 Use double precision for multiply and accumulate
-fp32 Use single precision for multiply and accumulate
-fp16 Use half precision for multiply and accumulate
-fp16fast Use half precision for multiply and single precision for accumulate
-tc Use Tensor cores (if available)
-l List all GPUs in the system
-i N Execute only on GPU N
-h Show this help message
Example:
gpu_burn -fp16fast -d 3600