Runtime Parameters: GPU - hyschive/gamer-fork GitHub Wiki

Parameters described on this page: OPT__GPUID_SELECT,   FLU_GPU_NPGROUP,   POT_GPU_NPGROUP,   CHE_GPU_NPGROUP,   GPU_NSTREAM  

Parameters below are shown in the format:   Name   (Valid Values)   [Default Value]

• OPT__GPUID_SELECT   (-2=CUDA, -1=MPI rank, ≥0=input)   [-1]

  • Description: See Set and Validate GPU IDs.
  • Restriction: Must be smaller than the total number of GPUs in a node.

• FLU_GPU_NPGROUP   (>0; ≤0 → set to default)   [depend on the GPU spec]

  • Description: Number of patch groups updated by the GPU/CPU fluid solvers at a single time. See also Performance Optimizations: GPU.
  • Restriction: Must be a multiple of GPU_NSTREAM.

• POT_GPU_NPGROUP   (>0; ≤0 → set to default)   [depend on the GPU spec]

  • Description: Number of patch groups updated by the GPU/CPU Poisson solvers at a single time. See also Performance Optimizations: GPU.
  • Restriction: Must be a multiple of GPU_NSTREAM.

• CHE_GPU_NPGROUP   (>0; ≤0 → set to default)   [depend on the GPU spec]

  • Description: Number of patch groups updated by the GPU/CPU GRACKLE solvers at a single time. See also Performance Optimizations: GPU. The GPU version is currently not supported.
  • Restriction:

• GPU_NSTREAM   (>0; ≤0 → set to default)   [depend on the GPU spec]

  • Description: Number of CUDA streams for the asynchronous memory copy between CPU and GPU. See also Performance Optimizations: GPU.
  • Restriction: See the restrictions on FLU_GPU_NPGROUP and POT_GPU_NPGROUP.

• Main page of Runtime Parameters
• Main page of GPU