Skip to content

UR CUDA CI machines have "Driver/library version mismatch" problem #17614

Open
@againull

Description

@againull

Describe the bug

NVML library is failing at initialization on CI UR CUDA machines because of the "Driver/library version mismatch" problem which is caused by incompatible versions on the systems.

As a result, currently, nvidia-smi output (in "Get information about platform" section) on any run is this:

**********CUDA Version**********
Failed to initialize NVML: Driver/library version mismatch
NVML library version: 550.144

Problem is visible on any UR CUDA run, for example, see "Get information about platform" section here:
https://github.com/intel/llvm/actions/runs/14037080040/job/39298376176?pr=17605

Please help to fix driver/library installations on machines.
Noticed it when trying to implement couple of descriptors via nvml in #17590.

To reproduce

To reproduce, you can just run nvidia-smi on the machine.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingcudaCUDA back-end

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions