HPL CUDA don't use GPGPU

Dear All,
I have successfully compile HPL CUDA downloaded from NVIDIA site. My server use single GPGPU (M2075). When I run ./run_linpack first time I get error not enough GPGPU. I change P and Q value in HPL.dat to 1. I can successfully run ./run_linpack afterward.

When I check top, the CPU process is 100%. but using : nvidia-smi -q -g 0 -d UTILIZATION -l
GPGPU, CPU and memory usage is 0%. I would like to ask :

  • Can I run HPL CUDA with 1 GPGPU per host?
  • If I run HPL CUDA will the program use both CPU and GPGPU? or it will only use CPU?
  • What is the correct way to use One GPGPU beside editing P & Q to 1 in HPL.dat?

Thank you all.

Hi sumodirjo,

The HPL CUDA from the NVIDIA site uses both CPU and GPU, you can set the number of threads used by CPU per GPU in the run_linpack script, the default value is 4, set it to 1.

PxQ=number of GPUs you have, so if you have only one GPU, set both for 1.

Hope this woks out for you, tell me if it does.