This repo contains performance benchmark on UNet and UNet3D, simply use ./run.sh
to launch the script.
On a CUDA available machine,benchmark.py
will run on GPU by default, need to specify --no-cuda
.
The script runs on CPU in case CUDA is not available, with the following setting
# set OMP_NUM_THREADS to be the number of physical cores
export KMP_AFFINITY="granularity=fine,compact,1,0"
export OMP_NUM_THREADS=56
export KMP_BLOCKTIM=1