Skip to content

Issues: NVIDIA/nccl

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

726 Open 736 Closed
726 Open 736 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Assignee
Filter by who’s assigned
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Issues list

Recent DGX 25.01 cause huge performance regression for NCCL <= 2.22.3
#1676 opened Apr 8, 2025 by xman1979
3
using avx to accelerate inter node protoll
#1675 opened Apr 7, 2025 by visualxu
nvidia kvm pcie passthrough how topo is configured for virtual machines
#1674 opened Apr 7, 2025 by Vagony
1
transport/nvls.cc:598 NCCL WARN Cuda failure 1 'invalid argument'
#1672 opened Apr 6, 2025 by cassanof
1
Is it possible to also avoid cross rail when channel count is odd?
#1671 opened Apr 5, 2025 by fishautumn
Identifying the GPU to NIC mapping used by NCCL via nccl logs question triaged
#1667 opened Apr 2, 2025 by pradeep-hk
7
Unexpected delay in Concurrent Groups with Different Rank Orders question triaged
#1666 opened Apr 1, 2025 by amjal
1
Kind Query about Why NCCL produce ".o" file for each datatype, instead of pass hyperparameters, e.g. "Datatype t" question
#1665 opened Apr 1, 2025 by CVPR-2020
NET/IBP:Got completion from peer *.*.*.*<***> with status=** opcode=***** len=***** vendor err 129 (Send) more info needed question
#1664 opened Apr 1, 2025 by limiao66
5
NCCL data race question triaged
#1663 opened Mar 27, 2025 by whll27182818
4
Segmentation fault while running all_reduce_perf 5090 bug more info needed
#1660 opened Mar 25, 2025 by S-o-T
2
NCCL profiler proxyStep start/stop counts mismatch in plugin v3. more info needed question
#1658 opened Mar 24, 2025 by jingxu9x
2
ncclInternalError: Internal check failed when using irecv with non-contiguous tensor in NCCL backend more info needed question
#1655 opened Mar 23, 2025 by default1360
1
ncclInternalError: Internal check failed and CUDA error: invalid device ordinal when using dist.recv_object_list with NCCL backend in multi-GPU setup more info needed question
#1654 opened Mar 23, 2025 by default1360
1
Very long initialization time with NCCL-GPUDirectTCPX question
#1652 opened Mar 21, 2025 by karakusc
1
[Feature Request] Allow specifying nccl{Send,Recv} peer via CUDA tensor enhancement
#1648 opened Mar 19, 2025 by garrett361
2
ncclCommInitRankConfig taking 60 seconds+ for DGX B200s more info needed question triaged
#1646 opened Mar 19, 2025 by isidentical
2
How data is moving between GPUs? question
#1644 opened Mar 19, 2025 by eliird
3
Add op status in RAS when comm is not ok question triaged
#1642 opened Mar 18, 2025 by gangxie112
1
does st.volatile.global.v4.u32 instruction atomic that LL protocol use question
#1641 opened Mar 18, 2025 by telala
3
Modeling NCCL Performance for AllToAll Communications in Various Topologies question
#1640 opened Mar 17, 2025 by sabuj7177
The performance of NCCL All-Reduce across 8 NVIDIA 5090 GPUs is extremely poor question triaged
#1639 opened Mar 14, 2025 by wowlizw1
1
Nccl stuck when pods are with istio envoy triaged
#1638 opened Mar 12, 2025 by ran-haim
10
NCCL P2P issue using two RTX 5090 bug duplicate fixed
#1637 opened Mar 11, 2025 by shahizat
14
Question about NCCL Ring Generation question
#1633 opened Mar 7, 2025 by WoosungMyung
ProTip! Exclude everything labeled bug with -label:bug.