10000 NCCL init hits CUDA failure 'invalid argument' on 12.2 driver · Issue #150852 · pytorch/pytorch · GitHub
[go: up one dir, main page]
Skip to content

NCCL init hits CUDA failure 'invalid argument' on 12.2 driver #150852

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

Jump to bottom
Open
kwen2501 opened this issue Apr 8, 2025 · 9 comments
Open

NCCL init hits CUDA failure 'invalid argument' on 12.2 driver #150852

kwen2501 opened this issue Apr 8, 2025 · 9 comments
Labels
has workaround module: nccl Problems related to nccl support oncall: distributed Add this issue/PR to distributed oncall triage queue triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module

Comments

@kwen2501
Copy link
Contributor
kwen2501 commented Apr 8, 2025 •
edited
Loading

🐛 Describe the bug

Error seen with nightly build, e.g. torch==2.8.0.dev20250327+cu126

[2025-04-08 08:39:46] devgpu263:589012:591652 [0] transport/nvls.cc:254 NCCL WARN Cuda failure 1 'invalid argument'
devgpu263:589012:591652 [0] NCCL INFO transport/nvls.cc:409 -> 1
devgpu263:589012:591652 [0] NCCL INFO init.cc:1141 -> 1
devgpu263:589012:591652 [0] NCCL INFO init.cc:1409 -> 1
devgpu263:589012:591652 [0] NCCL INFO group.cc:75 -> 1 [Async thread]
devgpu263:589012:589012 [0] NCCL INFO group.cc:422 -> 1
devgpu263:589012:589012 [0] NCCL INFO group.cc:581 -> 1
devgpu263:589012:589012 [0] NCCL INFO init.cc:1836 -> 1
[rank0]: Traceback (most recent call last):
[rank0]:   File "/home/kw2501/local/driver_12.2/repro.py", line 12, in <module>
[rank0]:     dist.all_reduce(x)
[rank0]:   File "/home/kw2501/.conda/envs/titan/lib/python3.10/site-packages/torch/distributed/c10d_logger.py", line 81, in wrapper
[rank0]:     return func(*args, **kwargs)
[rank0]:   File "/home/kw2501/.conda/envs/titan/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 2868, in all_reduce
[rank0]:     work = group.allreduce([tensor], opts)
[rank0]: torch.distributed.DistBackendError: NCCL error in: /pytorch/torch/csrc/distributed/c10d/NCCLUtils.cpp:77, unhandled cuda error (run with NCCL_DEBUG=INFO for details), NCCL version 2.26.2
[rank0]: ncclUnhandledCudaError: Call to CUDA function failed.
[rank0]: Last error:
[rank0]: Cuda failure 1 'invalid argument'

Mini repro:

import os
import torch
import torch.distributed as dist


if __name__ == "__main__":
    rank = int(os.getenv("RANK"))
    world_size = int(os.getenv("WORLD_SIZE"))
    device = torch.device("cuda", rank)
    dist.init_process_group("nccl", rank=rank, world_size=world_size)
    x = torch.empty(4, device=device)
    dist.all_reduce(x)
    print(x)

Command line:

NCCL_DEBUG=INFO torchrun --standalone --nproc-per-node 4 repro.py

Fails with 12.2 driver:
Driver Version: 535.154.05 CUDA Version: 12.2

Works with 12.4 driver:
Driver Version: 550.90.07 CUDA Version: 12.4

                 |   Run w/ 12.2 driver   |   Run w/ 12.4 driver or higher
torch Built w/   |   
12.2 toolkit     |          Works         |     Not tested
--------------------------------------------------------------------
torch Built w/   |   
12.6 toolkit     |          Fails         |          Works

Line 254 in nvls.cc:

CUCHECKGOTO(cuMulticastBindMem(*mcHandle, 0/*mcOffset*/, *ucHandle, 0/*memOffset*/, ucsize, 0/*flags*/), ret, fail);

Versions

[kw2501@devgpu263.prn2 ~]$ python collect_env.py 
Collecting environment information...
PyTorch version: 2.8.0.dev20250327+cu126
Is debug build: False
CUDA used to build PyTorch: 12.6
ROCM used to build PyTorch: N/A

OS: CentOS Stream 9 (x86_64)
GCC version: (GCC) 11.5.0 20240719 (Red Hat 11.5.0-5)
Clang version: Could not collect
CMake version: version 3.26.5
Libc version: glibc-2.34

Python version: 3.10.16 (main, Dec 11 2024, 16:24:50) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-6.4.3-0_fbk20_zion_2830_g3e5ab162667d-x86_64-with-glibc2.34
Is CUDA available: True
CUDA runtime version: 12.6.85
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: NVIDIA H100
GPU 1: NVIDIA H100
GPU 2: NVIDIA H100
GPU 3: NVIDIA H100
GPU 4: NVIDIA H100
GPU 5: NVIDIA H100
GPU 6: NVIDIA H100
GPU 7: NVIDIA H100

Nvidia driver version: 535.154.05
cuDNN version: Probably one of the following:
/usr/lib64/libcudnn.so.8.8.0
/usr/lib64/libcudnn.so.9.6.0
/usr/lib64/libcudnn_adv.so.9.6.0
/usr/lib64/libcudnn_adv_infer.so.8.8.0
/usr/lib64/libcudnn_adv_train.so.8.8.0
/usr/lib64/libcudnn_cnn.so.9.6.0
/usr/lib64/libcudnn_cnn_infer.so.8.8.0
/usr/lib64/libcudnn_cnn_train.so.8.8.0
/usr/lib64/libcudnn_engines_precompiled.so.9.6.0
/usr/lib64/libcudnn_engines_runtime_compiled.so.9.6.0
/usr/lib64/libcudnn_graph.so.9.6.0
/usr/lib64/libcudnn_heuristic.so.9.6.0
/usr/lib64/libcudnn_ops.so.9.6.0
/usr/lib64/libcudnn_ops_infer.so.8.8.0
/usr/lib64/libcudnn_ops_train.so.8.8.0
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                       x86_64
CPU op-mode(s):                     32-bit, 64-bit
Address sizes:                      52 bits physical, 57 bits virtual
Byte Order:                         Little Endian
CPU(s):                             384
On-line CPU(s) list:                0-383
Vendor ID:                          AuthenticAMD
Model name:                         AMD EPYC 9654 96-Core Processor
CPU family:                         25
Model:                              17
Thread(s) per core:                 2
Core(s) per socket:                 96
Socket(s):                          2
Stepping:                           1
Frequency boost:                    enabled
CPU(s) scaling MHz:                 84%
CPU max MHz:                        3707.8120
CPU min MHz:                        1500.0000
BogoMIPS:                           4792.81
Flags:                              fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid overflow_recov succor smca fsrm flush_l1d
Virtualization:                     AMD-V
L1d cache:                          6 MiB (192 instances)
L1i cache:                          6 MiB (192 instances)
L2 cache:                           192 MiB (192 instances)
L3 cache:                           768 MiB (24 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-95,192-287
NUMA node1 CPU(s):                  96-191,288-383
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit:        Not affected
Vulnerability L1tf:                 Not affected
Vulnerability Mds:                  Not affected
Vulnerability Meltdown:             Not affected
Vulnerability Mmio stale data:      Not affected
Vulnerability Retbleed:             Not affected
Vulnerability Spec store bypass:    Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:           Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:           Vulnerable: eIBRS with unprivileged eBPF
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Not affected

Versions of relevant libraries:
[pip3] numpy==2.1.2
[pip3] nvidia-cublas-cu12==12.6.4.1
[pip3] nvidia-cuda-cupti-cu12==12.6.80
[pip3] nvidia-cuda-nvrtc-cu12==12.6.77
[pip3] nvidia-cuda-runtime-cu12==12.6.77
[pip3] nvidia-cudnn-cu12==9.5.1.17
[pip3] nvidia-cufft-cu12==11.3.0.4
[pip3] nvidia-curand-cu12==10.3.7.77
[pip3] nvidia-cusolver-cu12==11.7.1.2
[pip3] nvidia-cusparse-cu12==12.5.4.2
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.6.85
[pip3] nvidia-nvtx-cu12==12.6.77
[pip3] pytorch-triton==3.3.0+git96316ce5
[pip3] torch==2.8.0.dev20250327+cu126
[pip3] torchaudio==2.6.0.dev20250329+cu126
[pip3] torchdata==0.11.0
[pip3] torchtitan==0.0.2
[pip3] torchvision==0.22.0.dev20250329+cu126
[pip3] triton==3.2.0
[conda] numpy                     2.1.2                    pypi_0    pypi
[conda] nvidia-cublas-cu12        12.6.4.1                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.6.80                  pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.6.77                  pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.6.77                  pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.5.1.17                 pypi_0    pypi
[conda] nvidia-cufft-cu12         11.3.0.4                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.7.77                pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.7.1.2                 pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.5.4.2                 pypi_0    pypi
[conda] nvidia-cusparselt-cu12    0.6.3                    pypi_0    pypi
[conda] nvidia-nccl-cu12          2.26.2                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.6.85                  pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.6.77                  pypi_0    pypi
[conda] pytorch-triton            3.3.0+git96316ce5          pypi_0    pypi
[conda] torch                     2.8.0.dev20250327+cu126          pypi_0    pypi
[conda] torchaudio                2.6.0.dev20250329+cu126
8000
          pypi_0    pypi
[conda] torchdata                 0.11.0                   pypi_0    pypi
[conda] torchtitan                0.0.2                    pypi_0    pypi
[conda] torchvision               0.22.0.dev20250329+cu126          pypi_0    pypi
[conda] triton                    3.2.0                    pypi_0    pypi

cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k
@kwen2501 kwen2501 pinned this issue Apr 8, 2025
@kwen2501 kwen2501 unpinned this issue Apr 8, 2025
@atalman
Copy link
Contributor
atalman commented Apr 8, 2025

cc @ptrblck @tinglvv @nWEIdia

@malfet malfet added oncall: distributed Add this issue/PR to distributed oncall triage queue module: nccl Problems related to nccl support labels Apr 8, 2025
@atalman atalman added this to the 2.7.0 milestone Apr 8, 2025
@ptrblck
Copy link
Collaborator
ptrblck commented Apr 8, 2025

@eqy

@kiskra-nvidia
Copy link

Is it possible to get a complete NCCL_DEBUG=INFO output from that failed run? For a good measure, maybe also with NCCL_DEBUG_SUBSYS=ALL?

@kwen2501
Copy link
Contributor Author
kwen2501 commented Apr 9, 2025 •
edited
Loading

Hi @kiskra-nvidia thanks! I added the following Pass/Fail matrix:

                 |   Run w/ 12.2 driver   |   Run w/ 12.4 driver or higher
torch Built w/   |   
12.2 toolkit     |          Works         |     Not tested
--------------------------------------------------------------------
torch Built w/   |   
12.6 toolkit     |          Fails         |          Works

@kwen2501
Copy link
Contributor Author
kwen2501 commented Apr 9, 2025

@kiskra-nvidia here is the INFO log with SUBSYS=ALL
https://gist.github.com/kwen2501/215f949b0d201a0ca2b9c907c2ec502a

It is a 8 x H100 machine with NVSwitch. I ran the test on 4 GPUs.

@kiskra-nvidia
Copy link

Thanks! No obvious clues in these logs, unfortunately.

Sometimes we see weird NVLS errors like this if the devices were not (re-)initialized in the right order (such as if the fabric manager is restarted without restarting the GPUs). Can you make sure that the initialization follows the order specified in NVIDIA Fabric Manager, section 4.3?

Also, when you say that it works with one driver version but not another, is that on the same node?

@kwen2501
Copy link
Contributor Author

Also, when you say that it works with one driver version but not another, is that on the same node?

No they are on different nodes.

@atalman atalman modified the milestones: 2.7.0, 2.7.1 Apr 23, 2025
@fduwjj fduwjj added the triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module label Apr 23, 2025
@malfet malfet removed this from the 2.7.1 milestone May 15, 2025
@malfet
Copy link
Contributor
malfet commented May 15, 2025

Demilestoning the issue as it has a workaround and there aren't much could be done on the PyTorch side

@tinglvv
Copy link
Collaborator
tinglvv commented May 15, 2025

Hi this should be fixed now with the latest nightlies with NCCL 2.26.5, @kwen2501 would you mind recheck if this issue persists? Thanks

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
has workaround module: nccl Problems related to nccl support oncall: distributed Add this issue/PR to distributed oncall triage queue triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
7 participants
@malfet @kwen2501 @fduwjj @atalman @ptrblck @tinglvv @kiskra-nvidia
0