Full documentation for RCCL is available at https://rccl.readthedocs.io
- Experimental support for clique-based kernels (opt in with RCCL_ENABLE_CLIQUE=1)
- Clique-based kernels may offer better performance for smaller input sizes
- Clique-based kernels are currently only enabled for AllReduce under a certain byte limit (controlled via RCCL_CLIQUE_ALLREDUCE_BYTE_LIMIT)
- Performance improvements for Rome-based systems
- Clique-based kernels are currently experimental and have not been fully tested on all topologies. By default, clique-based kernels are disabled if the detected topology is not supported (override with RCCL_FORCE_ENABLE_CLIQUE)
- Clique-based kernels may hang if there are differences between environment variables set across ranks.
- Clique-based kernels may fail if the input / output device pointers are not the base device pointers returned by hipMalloc.
- Adding support for alltoallv RCCL kernel
- Modifications to topology based on XGMI links
- None
- Support for static library builds
- None
- Updated to RCCL API version of 2.7.6
- Added gather, scatter and all-to-all collectives
- Updated to RCCL API version of 2.6.4
- Compatibility with NCCL 2.6
- Network interface improvements with API v3
- Fixing issues and built time improvements for hip-clang
- Network topology detection
- Improved CPU type detection
- Infiniband adaptive routing support
- Switched to hip-clang as default compiler
- Deprecated hcc build