Peter Boyle 5a9056cd93 Accelerator: lower default accelerator_threads from 16 to 8 ...

Benchmark_dwf_fp32 on MI250X GCD: 1.7 TF/s at nt=8, ~300 GF/s at nt=16.
With Nsimd=8 (fp32, GEN_SIMD_WIDTH=64B), nt=8 gives exactly 64 threads =
one full AMD wavefront. Higher values double register demand per block and
hit a register-pressure cliff for stencil kernels.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-05-21 12:34:30 -04:00

..

communication-overlap.md

skills: HPC battle-hardening skill files for GPU+MPI correctness

2026-05-21 12:34:30 -04:00

compiler-validation.md

skills: HPC battle-hardening skill files for GPU+MPI correctness

2026-05-21 12:34:30 -04:00

correctness-verification.md

skills: HPC battle-hardening skill files for GPU+MPI correctness

2026-05-21 12:34:30 -04:00

gpu-memory-performance.md

Accelerator: lower default accelerator_threads from 16 to 8

2026-05-21 12:34:30 -04:00

gpu-runtime-correctness.md

skills: HPC battle-hardening skill files for GPU+MPI correctness

2026-05-21 12:34:30 -04:00

hang-diagnosis.md

skills: HPC battle-hardening skill files for GPU+MPI correctness

2026-05-21 12:34:30 -04:00

mpi-heterogeneous.md

skills: HPC battle-hardening skill files for GPU+MPI correctness

2026-05-21 12:34:30 -04:00