PyTorch 2.15 released with rewritten NCCL backend as default
Overview
The PyTorch Foundation releases PyTorch 2.15, which makes the rewritten NCCL backend the default for every nccl process group, according to the PyTorch Blog.
Setting TORCH_DIST_USE_NCCL2=0 or using the nccl-legacy backend name restores the previous behavior.
The release also adds a flight recorder that dumps on hangs, FSDP2 precision and CUDA graph controls, and FP8 and decode speedups on Apple Silicon. Python 3.10 and macOS 14 support are dropped, and the default CUDA wheel moves to 13.2.
Written by AI from the articles below · updated Oct 28, 2:01 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- PyTorch BlogOfficialPickPyTorch 2.15 makes the rewritten NCCL backend the default
AIPyTorch 2.15 is released with a rewritten NCCL backend that is now the default for every nccl process group, with TORCH_DIST_USE_NCCL2=0 and nccl-legacy as rollback paths. The release also adds a flight recorder that dumps on hangs, FSDP2 precision and CUDA graph controls, and FP8 and decode speedups on Apple Silicon. Python 3.10 and macOS 14 support are dropped, and the default CUDA wheel moves to 13.2.
Heat Trend
- Current comparable heat
- 9
- Peak comparable heat
- 10Oct 28, 2:00 PM ET
- Change in past 24 hours
- –
The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.