Resolve CUDA performance regression

- Tried OpenMPI vs MPICH in several version flavors.
- Tried Docker "to reproduce exactly".
- Tried Spack and CUDA 11.9.
- Related to #4428.
This commit is contained in:
Hans Pabst 2025-10-02 10:21:20 +02:00
parent fdff660ab3
commit b396de2029
10 changed files with 11 additions and 11 deletions

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_cuda_A100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_cuda_P100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_cuda_V100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_hip_cuda_A100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_hip_cuda_P100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_hip_cuda_V100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_performance_cuda_A100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_performance_cuda_P100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -3,7 +3,7 @@
# Usage: podman build --shm-size=1g -f ./Dockerfile.test_performance_cuda_V100 ../../
#
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda

View file

@ -597,7 +597,7 @@ RUN make -j ARCH=Linux-x86-64-nvhpc VERSION=ssmp cp2k
# ======================================================================================
def install_deps_toolchain_cuda(gpu_ver: str) -> str:
return rf"""
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda
@ -626,7 +626,7 @@ RUN apt-get update -qq && apt-get install -qq --no-install-recommends \
# ======================================================================================
def install_deps_toolchain_hip_cuda(gpu_ver: str) -> str:
return rf"""
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
FROM nvidia/cuda:13.0.1-devel-ubuntu24.04
# Setup CUDA environment.
ENV CUDA_PATH /usr/local/cuda