deepseek-ai/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP.git synced 2026-11-05 00:47:32 +00:00
Code Issues 308 Projects Packages Wiki Activity

[GH-ISSUE #41] Is it normal that NCCL warnings and aborts always occur after running test_intranode.py? #39

New issue
Closed
opened 2026-10-26 09:53:05 +00:00 by bsakaguchi · 4 comments
bsakaguchi commented 2026-10-26 09:53:05 +00:00
Owner
Copy link

Originally created by @MinhuiWan on GitHub (Mar 4, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/41

When running test_intranode.py on an 8-H100 GPU machine， the occurrence of NCCL warnings and aborts is not normals. why? Is it Normal?

[testing] Running with BF16, without top-k (async=False, previous=False) ... passed
[testing] Running with BF16, with top-k (async=False, previous=False) ... passed
[testing] Running with BF16, without top-k (async=False, previous=False) ... passed
[testing] Running with BF16, with top-k (async=False, previous=False) ... passed
[testing] Running with FP8, without top-k (async=False, previous=False) ... passed
[testing] Running with FP8, with top-k (async=False, previous=False) ... passed
[testing] Running with BF16, without top-k (async=True, previous=False) ... passed
[testing] Running with BF16, with top-k (async=True, previous=False) ... passed
[testing] Running with BF16, without top-k (async=True, previous=False) ... passed
[testing] Running with BF16, with top-k (async=True, previous=False) ... passed
[testing] Running with FP8, without top-k (async=True, previous=False) ... passed
[testing] Running with FP8, with top-k (async=True, previous=False) ... passed
[testing] Running with BF16, without top-k (async=False, previous=True) ... passed
[testing] Running with BF16, with top-k (async=False, previous=True) ... passed
[testing] Running with BF16, without top-k (async=False, previous=True) ... passed
[testing] Running with BF16, with top-k (async=False, previous=True) ... passed
[testing] Running with FP8, without top-k (async=False, previous=True) ... passed
[testing] Running with FP8, with top-k (async=False, previous=True) ... passed
[testing] Running with BF16, without top-k (async=True, previous=True) ... passed
[testing] Running with BF16, with top-k (async=True, previous=True) ... passed
[testing] Running with BF16, without top-k (async=True, previous=True) ... passed
[testing] Running with BF16, with top-k (async=True, previous=True) ... passed
[testing] Running with FP8, without top-k (async=True, previous=True) ... passed
[testing] Running with FP8, with top-k (async=True, previous=True) ... passed

[tuning] SMs 24, NVL chunk 4: 257.81 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8: 282.84 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12: 266.20 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16: 256.93 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20: 253.48 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24: 250.78 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28: 247.80 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32: 245.42 GB/s (NVL)
[tuning] Best dispatch (FP8): SMs 24, NVL chunk 8, 282.84 GB/s (NVL)

[tuning] SMs 24, NVL chunk 4: 297.71 GB/s (NVL)
[tuning] SMs 24, NVL chunk 8: 266.76 GB/s (NVL)
[tuning] SMs 24, NVL chunk 12: 259.17 GB/s (NVL)
[tuning] SMs 24, NVL chunk 16: 252.50 GB/s (NVL)
[tuning] SMs 24, NVL chunk 20: 245.52 GB/s (NVL)
[tuning] SMs 24, NVL chunk 24: 245.22 GB/s (NVL)
[tuning] SMs 24, NVL chunk 28: 239.93 GB/s (NVL)
[tuning] SMs 24, NVL chunk 32: 237.85 GB/s (NVL)
[tuning] Best dispatch (BF16): SMs 24, NVL chunk 4, 297.71 GB/s (NVL)

[tuning] SMs 24, NVL chunk 1: 157.31 GB/s (NVL)
[tuning] SMs 24, NVL chunk 2: 295.94 GB/s (NVL)
[tuning] SMs 24, NVL chunk 3: 332.80 GB/s (NVL)
[tuning] SMs 24, NVL chunk 4: 335.94 GB/s (NVL)
[tuning] Best combine: SMs 24, NVL chunk 4: 335.94 GB/s (NVL)

0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:550 -> 3
0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:573 -> 3
0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:621 -> 3
0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:58 -> 3
0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:775 -> 3
0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:752 -> 3
0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:428 -> 3
0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:564 -> 3
0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:668 -> 3

0bc8f5b83646:2377:2986 [3] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable
0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:826 -> 3

0bc8f5b83646:2377:2986 [3] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 3, res=3, closed=0

0bc8f5b83646:2377:2986 [3] proxy.cc:1521 NCCL WARN [Proxy Service 3] Failed to execute operation Close from rank 3, retcode 3
[rank0]:[W304 06:59:09.530919875 ProcessGroupNCCL.cpp:1187] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present, but this warning has only been added since PyTorch 2.4 (function operator())
0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:550 -> 3
0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:573 -> 3
0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:621 -> 3
0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:58 -> 3
0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:775 -> 3
0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:752 -> 3
0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:428 -> 3
0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:564 -> 3
0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:668 -> 3

0bc8f5b83646:2374:2994 [0] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable
0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:826 -> 3

0bc8f5b83646:2374:2994 [0] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 0, res=3, closed=0

0bc8f5b83646:2374:2994 [0] proxy.cc:1521 NCCL WARN [Proxy Service 0] Failed to execute operation Close from rank 0, retcode 3
0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:550 -> 3
0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:573 -> 3
0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:621 -> 3
0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:752 -> 3
0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:428 -> 3
0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:564 -> 3
0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:668 -> 3
0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:58 -> 3
0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:775 -> 3

0bc8f5b83646:2378:2984 [4] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable
0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:826 -> 3

0bc8f5b83646:2378:2984 [4] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 4, res=3, closed=0

0bc8f5b83646:2378:2984 [4] proxy.cc:1521 NCCL WARN [Proxy Service 4] Failed to execute operation Close from rank 4, retcode 3
0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:550 -> 3
0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:573 -> 3
0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:621 -> 3
0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:58 -> 3
0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:775 -> 3
0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:752 -> 3
0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:428 -> 3
0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:564 -> 3
0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:668 -> 3

0bc8f5b83646:2376:2996 [2] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable
0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:826 -> 3

0bc8f5b83646:2376:2996 [2] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 2, res=3, closed=0

0bc8f5b83646:2376:2996 [2] proxy.cc:1521 NCCL WARN [Proxy Service 2] Failed to execute operation Close from rank 2, retcode 3
0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:550 -> 3
0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:573 -> 3
0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:621 -> 3
0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:58 -> 3
0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:775 -> 3
0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:47 -> 3
0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:752 -> 3
0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:428 -> 3
0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:564 -> 3
0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:668 -> 3

0bc8f5b83646:2379:2998 [5] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable
0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:826 -> 3

0bc8f5b83646:2379:2998 [5] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 5, res=3, closed=0

0bc8f5b83646:2379:2998 [5] proxy.cc:1521 NCCL WARN [Proxy Service 5] Failed to execute operation Close from rank 5, retcode 3

Originally created by @MinhuiWan on GitHub (Mar 4, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/41 When running test_intranode.py on an 8-H100 GPU machine， the occurrence of NCCL warnings and aborts is not normals. why? Is it Normal? [testing] Running with BF16, without top-k (async=False, previous=False) ... passed [testing] Running with BF16, with top-k (async=False, previous=False) ... passed [testing] Running with BF16, without top-k (async=False, previous=False) ... passed [testing] Running with BF16, with top-k (async=False, previous=False) ... passed [testing] Running with FP8, without top-k (async=False, previous=False) ... passed [testing] Running with FP8, with top-k (async=False, previous=False) ... passed [testing] Running with BF16, without top-k (async=True, previous=False) ... passed [testing] Running with BF16, with top-k (async=True, previous=False) ... passed [testing] Running with BF16, without top-k (async=True, previous=False) ... passed [testing] Running with BF16, with top-k (async=True, previous=False) ... passed [testing] Running with FP8, without top-k (async=True, previous=False) ... passed [testing] Running with FP8, with top-k (async=True, previous=False) ... passed [testing] Running with BF16, without top-k (async=False, previous=True) ... passed [testing] Running with BF16, with top-k (async=False, previous=True) ... passed [testing] Running with BF16, without top-k (async=False, previous=True) ... passed [testing] Running with BF16, with top-k (async=False, previous=True) ... passed [testing] Running with FP8, without top-k (async=False, previous=True) ... passed [testing] Running with FP8, with top-k (async=False, previous=True) ... passed [testing] Running with BF16, without top-k (async=True, previous=True) ... passed [testing] Running with BF16, with top-k (async=True, previous=True) ... passed [testing] Running with BF16, without top-k (async=True, previous=True) ... passed [testing] Running with BF16, with top-k (async=True, previous=True) ... passed [testing] Running with FP8, without top-k (async=True, previous=True) ... passed [testing] Running with FP8, with top-k (async=True, previous=True) ... passed [tuning] SMs 24, NVL chunk 4: 257.81 GB/s (NVL) [tuning] SMs 24, NVL chunk 8: 282.84 GB/s (NVL) [tuning] SMs 24, NVL chunk 12: 266.20 GB/s (NVL) [tuning] SMs 24, NVL chunk 16: 256.93 GB/s (NVL) [tuning] SMs 24, NVL chunk 20: 253.48 GB/s (NVL) [tuning] SMs 24, NVL chunk 24: 250.78 GB/s (NVL) [tuning] SMs 24, NVL chunk 28: 247.80 GB/s (NVL) [tuning] SMs 24, NVL chunk 32: 245.42 GB/s (NVL) [tuning] Best dispatch (FP8): SMs 24, NVL chunk 8, 282.84 GB/s (NVL) [tuning] SMs 24, NVL chunk 4: 297.71 GB/s (NVL) [tuning] SMs 24, NVL chunk 8: 266.76 GB/s (NVL) [tuning] SMs 24, NVL chunk 12: 259.17 GB/s (NVL) [tuning] SMs 24, NVL chunk 16: 252.50 GB/s (NVL) [tuning] SMs 24, NVL chunk 20: 245.52 GB/s (NVL) [tuning] SMs 24, NVL chunk 24: 245.22 GB/s (NVL) [tuning] SMs 24, NVL chunk 28: 239.93 GB/s (NVL) [tuning] SMs 24, NVL chunk 32: 237.85 GB/s (NVL) [tuning] Best dispatch (BF16): SMs 24, NVL chunk 4, 297.71 GB/s (NVL) [tuning] SMs 24, NVL chunk 1: 157.31 GB/s (NVL) [tuning] SMs 24, NVL chunk 2: 295.94 GB/s (NVL) [tuning] SMs 24, NVL chunk 3: 332.80 GB/s (NVL) [tuning] SMs 24, NVL chunk 4: 335.94 GB/s (NVL) [tuning] Best combine: SMs 24, NVL chunk 4: 335.94 GB/s (NVL) 0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:550 -> 3 0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:573 -> 3 0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:621 -> 3 0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:58 -> 3 0bc8f5b83646:2377:3009 [3] NCCL INFO misc/socket.cc:775 -> 3 0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:752 -> 3 0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:428 -> 3 0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:564 -> 3 0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:668 -> 3 0bc8f5b83646:2377:2986 [3] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable 0bc8f5b83646:2377:2986 [3] NCCL INFO misc/socket.cc:826 -> 3 0bc8f5b83646:2377:2986 [3] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 3, res=3, closed=0 0bc8f5b83646:2377:2986 [3] proxy.cc:1521 NCCL WARN [Proxy Service 3] Failed to execute operation Close from rank 3, retcode 3 [rank0]:[W304 06:59:09.530919875 ProcessGroupNCCL.cpp:1187] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present, but this warning has only been added since PyTorch 2.4 (function operator()) 0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:550 -> 3 0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:573 -> 3 0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:621 -> 3 0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:58 -> 3 0bc8f5b83646:2374:3011 [0] NCCL INFO misc/socket.cc:775 -> 3 0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:752 -> 3 0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:428 -> 3 0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:564 -> 3 0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:668 -> 3 0bc8f5b83646:2374:2994 [0] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable 0bc8f5b83646:2374:2994 [0] NCCL INFO misc/socket.cc:826 -> 3 0bc8f5b83646:2374:2994 [0] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 0, res=3, closed=0 0bc8f5b83646:2374:2994 [0] proxy.cc:1521 NCCL WARN [Proxy Service 0] Failed to execute operation Close from rank 0, retcode 3 0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:550 -> 3 0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:573 -> 3 0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:621 -> 3 0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:752 -> 3 0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:428 -> 3 0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:564 -> 3 0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:668 -> 3 0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:58 -> 3 0bc8f5b83646:2378:3013 [4] NCCL INFO misc/socket.cc:775 -> 3 0bc8f5b83646:2378:2984 [4] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable 0bc8f5b83646:2378:2984 [4] NCCL INFO misc/socket.cc:826 -> 3 0bc8f5b83646:2378:2984 [4] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 4, res=3, closed=0 0bc8f5b83646:2378:2984 [4] proxy.cc:1521 NCCL WARN [Proxy Service 4] Failed to execute operation Close from rank 4, retcode 3 0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:550 -> 3 0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:573 -> 3 0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:621 -> 3 0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:58 -> 3 0bc8f5b83646:2376:3015 [2] NCCL INFO misc/socket.cc:775 -> 3 0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:752 -> 3 0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:428 -> 3 0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:564 -> 3 0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:668 -> 3 0bc8f5b83646:2376:2996 [2] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable 0bc8f5b83646:2376:2996 [2] NCCL INFO misc/socket.cc:826 -> 3 0bc8f5b83646:2376:2996 [2] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 2, res=3, closed=0 0bc8f5b83646:2376:2996 [2] proxy.cc:1521 NCCL WARN [Proxy Service 2] Failed to execute operation Close from rank 2, retcode 3 0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:550 -> 3 0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:573 -> 3 0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:621 -> 3 0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:58 -> 3 0bc8f5b83646:2379:3017 [5] NCCL INFO misc/socket.cc:775 -> 3 0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:47 -> 3 0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:752 -> 3 0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:428 -> 3 0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:564 -> 3 0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:668 -> 3 0bc8f5b83646:2379:2998 [5] proxy.cc:1458 NCCL WARN [Service thread] Accept failed Resource temporarily unavailable 0bc8f5b83646:2379:2998 [5] NCCL INFO misc/socket.cc:826 -> 3 0bc8f5b83646:2379:2998 [5] proxy.cc:1497 NCCL WARN [Service thread] Could not receive type from localRank 5, res=3, closed=0 0bc8f5b83646:2379:2998 [5] proxy.cc:1521 NCCL WARN [Proxy Service 5] Failed to execute operation Close from rank 5, retcode 3
bsakaguchi commented 2026-10-26 09:53:06 +00:00
Author
Owner
Copy link

@sphish commented on GitHub (Mar 4, 2025):

No, NCCL is not necessary for DeepEP, you can disable NCCL by setting build env

NVSHMEM_USE_NCCL=0

See: https://github.com/deepseek-ai/DeepEP/tree/main/third-party#5-build-and-installation

<!-- gh-comment-id:2696915598 --> @sphish commented on GitHub (Mar 4, 2025): No, NCCL is not necessary for DeepEP, you can disable NCCL by setting build env ```bash NVSHMEM_USE_NCCL=0 ``` See: https://github.com/deepseek-ai/DeepEP/tree/main/third-party#5-build-and-installation
bsakaguchi commented 2026-10-26 09:53:07 +00:00
Author
Owner
Copy link

@MinhuiWan commented on GitHub (Mar 5, 2025):

No, NCCL is not necessary for DeepEP, you can disable NCCL by setting build env

NVSHMEM_USE_NCCL=0
See: https://github.com/deepseek-ai/DeepEP/tree/main/third-party#5-build-and-installation

I have set NVSHMEM_USE_NCCL=0 during install nvshmem， but execute “python test_intranode.py”， will call init_dist and use nccl。Will this have an impact?

<!-- gh-comment-id:2700037092 --> @MinhuiWan commented on GitHub (Mar 5, 2025): > No, NCCL is not necessary for DeepEP, you can disable NCCL by setting build env > > NVSHMEM_USE_NCCL=0 > See: https://github.com/deepseek-ai/DeepEP/tree/main/third-party#5-build-and-installation I have set NVSHMEM_USE_NCCL=0 during install nvshmem， but execute “python test_intranode.py”， will call init_dist and use nccl。Will this have an impact?
bsakaguchi commented 2026-10-26 09:53:07 +00:00
Author
Owner
Copy link

@sphish commented on GitHub (Mar 5, 2025):

This behavior doesn't meet expectations. You may need to check your network environment.

<!-- gh-comment-id:2700145620 --> @sphish commented on GitHub (Mar 5, 2025): This behavior doesn't meet expectations. You may need to check your network environment.
bsakaguchi commented 2026-10-26 09:53:07 +00:00
Author
Owner
Copy link

@MinhuiWan commented on GitHub (Mar 9, 2025):

This behavior doesn't meet expectations. You may need to check your network environment.
I set NCCL_DEBUG=INFO， so when run test_intranode.py, NCCL log is printed at the start and end，because of creating deep_ep.Buffer synchronizes NCCL device info. However, there are no NCCL logs during the dispatch and combine interface execute. Therefore, DeepEP's communication interfaces do not call NCCL.

<!-- gh-comment-id:2708662759 --> @MinhuiWan commented on GitHub (Mar 9, 2025): > This behavior doesn't meet expectations. You may need to check your network environment. I set NCCL_DEBUG=INFO， so when run test_intranode.py, NCCL log is printed at the start and end，because of creating deep_ep.Buffer synchronizes NCCL device info. However, there are no NCCL logs during the dispatch and combine interface execute. Therefore, DeepEP's communication interfaces do not call NCCL.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
hybrid-ep
chore/remove-clang-format-yapf
v2.5
nvDev
try-fix-cu128
mori-ep
antgroup-opt
compat-old-torch
add-linter
eager-rdma
v1.2.1
Labels
Clear labels   
pull-request
Mirrored from GitHub Pull Request

No labels
pull-request
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepEP#39
No description provided.