From 81cee7227c78be1cee0c3f4e8b1589ee040f551a Mon Sep 17 00:00:00 2001 From: Gabriele Bozzola Date: Sat, 22 Jun 2024 13:04:59 -0700 Subject: [PATCH] Fix GPU sharing on GPU-longruns --- .buildkite/longruns_gpu/pipeline.yml | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/.buildkite/longruns_gpu/pipeline.yml b/.buildkite/longruns_gpu/pipeline.yml index 0f3cde5f56..c91b7e2b7e 100644 --- a/.buildkite/longruns_gpu/pipeline.yml +++ b/.buildkite/longruns_gpu/pipeline.yml @@ -4,14 +4,13 @@ agents: modules: julia/1.10.0 cuda/julia-pref openmpi/4.1.5-mpitrampoline nsight-systems/2024.2.1 env: - JULIA_CUDA_MEMORY_POOL: none JULIA_MPI_HAS_CUDA: "true" JULIA_NVTX_CALLBACKS: gc JULIA_MAX_NUM_PRECOMPILE_FILES: 100 OPENBLAS_NUM_THREADS: 1 OMPI_MCA_opal_warn_on_missing_libcuda: 0 SLURM_KILL_BAD_EXIT: 1 - SLURM_GPU_BIND: none # https://github.com/open-mpi/ompi/issues/11949#issuecomment-1737712291 + SLURM_GRES_FLAGS: "allow-task-sharing" CONFIG_PATH: "config/longrun_configs" CLIMAATMOS_GC_NSTEPS: 10