BEAR Applications: NCCL 2.12.12-GCCcore-11.3.0-CUDA-11.7.0 (for BlueBEAR, BEARCloud VMs, and CaStLeS VMs)

NCCL 2.12.12-GCCcore-11.3.0-CUDA-11.7.0

The NVIDIA Collective Communications Library (NCCL) implements multi-GPU and multi-node collective communication primitives that are performance optimized for NVIDIA GPUs.

Accessing NCCL 2.12.12-GCCcore-11.3.0-CUDA-11.7.0

To load the module for NCCL 2.12.12-GCCcore-11.3.0-CUDA-11.7.0 please use this command on the BEAR systems (BlueBEAR, BEARCloud VMs, and CaStLeS VMs):

module load bear-apps/2021b module load NCCL/2.12.12-GCCcore-11.3.0-CUDA-11.7.0

BEAR Apps Version

2021b

More Information

For more information visit the NCCL website.

Dependencies

This version of NCCL has a direct dependency on: CUDA/11.7.0 GCCcore/11.3.0 UCX-CUDA/1.12.1-GCCcore-11.3.0-CUDA-11.7.0

Required By

This version of NCCL is a direct dependent of: AlphaFold/2.3.1-foss-2022a-CUDA-11.7.0 Horovod/0.28.0-foss-2022a-CUDA-11.7.0-PyTorch-2.0.1 jax/0.3.14-foss-2022a-CUDA-11.7.0 jax/0.4.4-foss-2022a-CUDA-11.7.0 PyTorch/2.0.1-foss-2022a-CUDA-11.7.0 PyTorch/1.12.1-foss-2022a-CUDA-11.7.0 TensorFlow/2.11.0-foss-2022a-CUDA-11.7.0 UCC-CUDA/1.0.0-GCCcore-11.3.0-CUDA-11.7.0

Other Versions

These versions of NCCL are available on the BEAR systems (BlueBEAR, BEARCloud VMs, and CaStLeS VMs). These will be retained in accordance with our Applications Support and Retention Policy.

Version	BEAR Apps Version
2.10.3-GCCcore-11.2.0-CUDA-11.4.1	2021b
2.10.3-GCCcore-10.3.0-CUDA-11.3.1	2021a
2.8.3-CUDA-11.1.1	2020b
2.7.8-CUDA-11.0.2	2020a
2.4.8-gcccuda-2019b	2019b
2.4.7-gcccuda-2019a-ppc64le	2019a
2.4.7-gcccuda-2019a	2019a
2.4.2-fosscuda-2018b	2018b

Last modified on 20th January 2023