NCCL
The NVIDIA Collective Communications Library (NCCL) implements multi-GPU and multi-node collectivecommunication primitives that are performance optimized for NVIDIA GPUs.
https://developer.nvidia.com/nccl
Available modules
The overview below shows which NCCL installations are available per target architecture in the HPCC module system, ordered based on software version (new to old).
To start using NCCL, load one of these modules using a module load command like:
1 | |
(This data was automatically generated on Thu, 15 May 2025 at 12:00:14 EDT)
| neoverse_v1 | generic | zen2 | zen3 | zen4 | haswell | skylake_avx512 | |
|---|---|---|---|---|---|---|---|
| Grace Nodes | everywhere (except Grace nodes) | amd20 | amd22 | amd24 | intel16 | intel18,amd20-v100,amd21,intel21 | |
| NCCL/2.20.5-GCCcore-13.2.0-CUDA-12.4.0 | - | - | x | x | x | - | x |
| NCCL/2.18.3-GCCcore-12.3.0-CUDA-12.1.1 | - | x | - | x | x | - | x |
| NCCL/2.12.12-GCCcore-11.3.0-CUDA-11.7.0 | - | - | - | - | - | x | - |