Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
Expand Up @@ -5,8 +5,8 @@ set -xe
rm -rf /shared/${1}

module load ${1}
NCCL_BENCHMARKS_VERSION='2.16.7'
NCCL_VERSION='2.27.7-1'
NCCL_BENCHMARKS_VERSION='2.17.1'
NCCL_VERSION='2.28.3-1'
MPI_HOME=$(which mpirun | awk -F '/bin' '{print $1}')
NVCC_GENCODE="-gencode=arch=compute_80,code=sm_80 -gencode=arch=compute_90,code=sm_90 -gencode=arch=compute_90,code=compute_90" # Arch for NVIDIA A100 and H100, ref https://docs.nvidia.com/cuda/ada-compatibility-guide/index.html

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -3,8 +3,8 @@
#SBATCH --exclusive

module load openmpi
NCCL_VERSION='2.27.7-1'
NCCL_BENCHMARKS_VERSION='2.16.7'
NCCL_VERSION='2.28.3-1'
NCCL_BENCHMARKS_VERSION='2.17.1'

. /etc/os-release
if [[ $ID==rhel || $ID==rocky ]]; then
Expand Down
4 changes: 2 additions & 2 deletions tests/integration-tests/tests/common/nccl_common.py
Original file line number Diff line number Diff line change
Expand Up @@ -69,8 +69,8 @@ def install_and_run_nccl_benchmarks(remote_command_executor, mpi_module, schedul
"p4d.24xlarge": 26.0,
# p5.48xlarge - Expected "in-place busbw" bandwidth with 2 nodes, 8 tasks per node is about 250GB/s
"p5.48xlarge": 250.0,
"p6-b200.48xlarge": 300,
"p6e-gb200.36xlarge": 500,
"p6-b200.48xlarge": 570, # Initial testing performance 631.17
"p6e-gb200.36xlarge": 650, # Initial testing performance 719.17
}

expected_bandwidth = instance_bandwidth_dict.get(instance)
Expand Down
Loading