Signal 8 (SIGFPE): a C/CUDA extension triggered a floating-point exception at runtime.

[CAIJN2021]

During the testing on NVIDIA H20 GPUs under a CUDA 13.0 environment, I encountered a floating-point exception (SIGFPE). The error log is as follows:

Traceback (most recent call last):
  File "/mnt/nfs_share/zhangqi/env/streamvln/bin/torchrun", line 7, in <module>
    sys.exit(main())
  File "/mnt/nfs_share/zhangqi/env/streamvln/lib/python3.9/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 346, in wrapper
    return f(*args, **kwargs)
  File "/mnt/nfs_share/zhangqi/env/streamvln/lib/python3.9/site-packages/torch/distributed/run.py", line 806, in main
    run(args)
  File "/mnt/nfs_share/zhangqi/env/streamvln/lib/python3.9/site-packages/torch/distributed/run.py", line 797, in run
    elastic_launch(
  File "/mnt/nfs_share/zhangqi/env/streamvln/lib/python3.9/site-packages/torch/distributed/launcher/api.py", line 134, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/mnt/nfs_share/zhangqi/env/streamvln/lib/python3.9/site-packages/torch/distributed/launcher/api.py", line 264, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
=======================================================
streamvln/streamvln_eval.py FAILED
-------------------------------------------------------
Failures:
[1]:
  time      : 2026-01-03_12:40:33
  host      : h20node6
  rank      : 1 (local_rank: 1)
  exitcode  : -8 (pid: 2935278)
  error_file: <N/A>
  traceback : Signal 8 (SIGFPE) received by PID 2935278
[2]:
  time      : 2026-01-03_12:40:33
  host      : h20node6
  rank      : 2 (local_rank: 2)
  exitcode  : -8 (pid: 2935279)
  error_file: <N/A>
  traceback : Signal 8 (SIGFPE) received by PID 2935279
[3]:
  time      : 2026-01-03_12:40:33
  host      : h20node6
  rank      : 3 (local_rank: 3)
  exitcode  : -8 (pid: 2935281)
  error_file: <N/A>
  traceback : Signal 8 (SIGFPE) received by PID 2935281
[4]:
  time      : 2026-01-03_12:40:33
  host      : h20node6
  rank      : 4 (local_rank: 4)
  exitcode  : -8 (pid: 2935282)
  error_file: <N/A>
  traceback : Signal 8 (SIGFPE) received by PID 2935282
[5]:
  time      : 2026-01-03_12:40:33
  host      : h20node6
  rank      : 5 (local_rank: 5)
  exitcode  : -8 (pid: 2935283)
  error_file: <N/A>
  traceback : Signal 8 (SIGFPE) received by PID 2935283
-------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2026-01-03_12:40:33
  host      : h20node6
  rank      : 0 (local_rank: 0)
  exitcode  : -8 (pid: 2935277)
  error_file: <N/A>
  traceback : Signal 8 (SIGFPE) received by PID 2935277
=======================================================

After switching to H800 GPUs with a lower CUDA version, the program runs normally. I’m not sure whether this issue is caused by the CUDA version.