Skip to content

Very slow inference on Linux 32-bit (CPU-only) vs fast performance on Win32/Win64/Linux64 on same hardware #3758

Description

@ds-kimi

I’m seeing a major performance issue specifically on Linux 32-bit (CPU-only).

  • Same processor (same machine)
  • Same model/configuration
  • CPU-only in all tests

Results:

  • Windows 32-bit: fast
  • Windows 64-bit: fast
  • Linux 64-bit: fast
  • Linux 32-bit: extremely slow

On Linux 32-bit, inference appears to hang for a long time.
If output eventually appears, it is often just:

!!!!!!!!!!!!!

Expected behavior:
Linux 32-bit may be slower, but it should still produce normal output and not stall this severely.

Can you confirm whether this is a known limitation/regression in the Linux x86 (32-bit) CPU path?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions