Search results

[Issue]: send/recv hangs when using NCCL_NCHANNELS_PER_NET_PEER=32

Status: Open.

#1915 In NVIDIA/nccl;

· zoran-zhu opened on Nov 19, 2025

[Issue]: torchrun segfaults with exit code -11. But passes when I disable NCCL_P2P and NCCL_SHM

Status: Open.

#1914 In NVIDIA/nccl;

· nishitnshah opened on Nov 18, 2025

NCCL 2.28.9 GIN Incompatible with External Network Plugins on Multi-Rail EFA

Status: Open.

#1913 In NVIDIA/nccl;

· dmvevents opened on Nov 18, 2025

[RFE]: NCCL RAS when mismatch print the list of stragglers

enhancement

Status: Open.

#1911 In NVIDIA/nccl;

· xman1979 opened on Nov 13, 2025

[Question]: Missing refCount Decrement for ncclProfileKernelLaunch?

question

Status: Open.

#1910 In NVIDIA/nccl;

· xxxxx-ctrl opened on Nov 13, 2025

[Question]: Why is collApiPoolBase incremented twice for the same event in profiler code?

question

Status: Open.

#1908 In NVIDIA/nccl;

· xxxxx-ctrl opened on Nov 12, 2025

[RFE]: reduce unnecessary cuda memory alloc

enhancement

Status: Open.

#1907 In NVIDIA/nccl;

· visualxu opened on Nov 10, 2025

[Issue]: <code>all_gather</code> returns tensors in incorrect rank order for certain permutation of <code>CUDA_VISIBLE_DEVICES</code> and <code>NCCL_ALGO=NVLS </code> (regression in NCCL 2.26.2; NCCL 2.21.5 OK)

Status: Open.

#1906 In NVIDIA/nccl;

· xiangxl-a opened on Nov 10, 2025

[Issue]: NCCL WARN communicator encountered a fatal error (detected in ncclIbTest)

Status: Open.

#1905 In NVIDIA/nccl;

· thqq479 opened on Nov 9, 2025

[Issue]: enable NCCL_GRAPH_MIXING_SUPPORT will launch a large number of stream in cudagraph

Status: Open.

#1901 In NVIDIA/nccl;

· MARD1NO opened on Nov 6, 2025

[Question]: Inspector: why exec time is the max of channel exec times?

question

Status: Open.

#1900 In NVIDIA/nccl;

· RicoloveFeng opened on Nov 5, 2025

[Question]: why PXN graph is prefered than crossNic ?

question

Status: Open.

#1899 In NVIDIA/nccl;

· fishautumn opened on Nov 5, 2025