Intel oneCCL 2021.14 Brings New Performance & Scalability Optimizations

Intel's oneCCL open-source software that is the oneAPI Collective Communications Library focused on providing an efficient implementation of deep learning communication patterns is out with a new release.

The oneAPI Collective Communications Library can be used with the likes of the PyTorch and Horovod frameworks and has been seeing renewed development now as part of the Unified Acceleration (UXL) Foundation efforts. With the oneCCL 2021.14 release there are optimizations for their key-value store to scale up to 3,000 nodes plus an assortment of other new performance optimizations:

- Optimizations on key-value store support to scale up to 3000 nodes
- New APIs for Allgather, Broadcast and group API calls
- Performance Optimizations for scaleup for Allgather, Allreduce, and Reduce-scatter for scaleup and scaleout
- Performance Optimizations for CPU single node
- Optimizations to reuse Level Zero events.
- Change of the default mechanism for IPC exchange to pidfd

Downloads and more details on the oneCCL 2021.14 release via GitHub.