Ultra and Unified CCL

ai allreduce amd broadcom collective cuda gpu hpc llm networking nvidia rdma
4 Open Issues Need Help Last updated: Jun 19, 2025

Open Issues Need Help

View All on GitHub
Make P2P work for EFA about 1 month ago
good first issue help wanted

Ultra and Unified CCL

C++
#ai#allreduce#amd#broadcom#collective#cuda#gpu#hpc#llm#networking#nvidia#rdma
good first issue

Ultra and Unified CCL

C++
#ai#allreduce#amd#broadcom#collective#cuda#gpu#hpc#llm#networking#nvidia#rdma

AI Summary: Create an `examples` directory within the UCCL repository containing multiple PyTorch examples demonstrating UCCL usage with both Nvidia and AMD GPUs. These examples should expand upon the existing `run_ddp.sh` script, providing more comprehensive and illustrative use cases.

Complexity: 4/5
good first issue

Ultra and Unified CCL

C++
#ai#allreduce#amd#broadcom#collective#cuda#gpu#hpc#llm#networking#nvidia#rdma

AI Summary: Modify the `misc/Makefile` in the UCCL project to remove the hard dependency on EFA (AWS Elastic Fabric Adapter) for miscellaneous builds. This involves adjusting the Makefile rules to conditionally skip builds that require EFA, ensuring that the miscellaneous components can be built even without EFA being present.

Complexity: 3/5
good first issue

Ultra and Unified CCL

C++
#ai#allreduce#amd#broadcom#collective#cuda#gpu#hpc#llm#networking#nvidia#rdma