Skip to content

vllm not starting on gfx1151 due to torch.OutOfMemoryError #5

Description

@ckuethe

Lemonade 10.4.0, Ubuntu 26.04, Linux 7.0.0-15, gfx1151... Freshly rebooted

2026-05-08 11:47:02.257 [Error] (Process) (EngineCore pid=2651) torch.OutOfMemoryError: HIP out of memory. Tried to allocate 12.95 GiB. GPU 0 has a total capacity of 127.00 GiB of which 18.55 GiB is free. Of the allocated memory 99.37 GiB is allocated by PyTorch, and 121.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
...
2026-05-08 11:47:02.257 [Error] (Process) (EngineCore pid=2651) ERROR 05-08 11:47:02 [core.py:1136] torch.OutOfMemoryError: HIP out of memory. Tried to allocate 12.95 GiB. GPU 0 has a total capacity of 127.00 GiB of which 18.55 GiB is free. Of the allocated memory 99.37 GiB is allocated by PyTorch, and 121.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
2026-05-08 11:47:03.123 [Info] (Process) [rank0]:[W508 11:47:03.157708266 ProcessGroupNCCL.cpp:1554] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())

Full log in next comment.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions