Hi,
I ran HPL benchmark with the system of 1 Xeon CPU E5-2650 v4 + 1 Tesla V100, compiled successfully with MKL + CUDA-9.0 + MPICH-3.2.
When I ran the test, I could not get any results because of that error.
The matrix size is pretty small (N=10000).
These are parameters I used:
export LD_LIBRARY_PATH= …
HPL_DIR=/…/hpl-2.0_FERMI_v15
export CPU_CORES_PER_GPU=20
export MKL_NUM_THREADS=$CPU_CORES_PER_GPU
export MKL_DYNAMIC=FALSE
export CUDA_DGEMM_SPLIT=0.66
export CUDA_DTRSM_SPLIT=0.56
export OMP_NUM_THREADS=$CPU_CORES_PER_GPU
Does anyone know what is the problem here? What should I do to fix it?
Many thanks,