NVIDIA HPC-X Software Toolkit

Unified Collective Communication (UCC)

Unified Collective Communication (UCC) was codesigned with industry partners for PyTorch-based deep learning recommender model training on multi-rail GPU platforms. UCC has been specifically designed and implemented for high-performance PGAS applications and runtimes. It serves as a replacement for HCOLL. UCC fully implements the range of HCOLL's hierarchical algorithms as well as full support for Nvidia's GPUs.     

For further information on what UCC is and how to use it, please see https://github.com/openucx/ucc

Please see UCC PyTorch integration layer, Torch_UCC at https://github.com/facebookresearch/torch_ucc
UCC is integrated into PyTorch via the UCC process group.

UCC is supported in both MPI and OSHMEM and enabled by default.

  • To toggle in MPI, set -mca coll_ucc_enable to 1/0.

  • To toggle in OSHMEM, set -mca coll_scoll_enable to 1/0.

  • You can additionally set collective frameworks via the  --mca coll
    eg. --mca coll ucc,basic,libnbc for OMPI or --mca scoll ucc,basic,libnbc for OSHMEM


Last updated: