microsoft / microsoft/WSL

CUDA fails to load

Open
#14,446 4 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

GPU
Dominant language
C++
Stars
33.7k
Forks
1.8k
Avg merge
3d 17h
Merged PRs (30d)
116

Description

Windows Version

10.0.26200.8037

WSL Version

2.6.3.0, 2.7.0.0

Are you using WSL 1 or WSL 2?
  • WSL 2
  • WSL 1
Kernel Version

6.6.87.2-1, 6.6.114.1-1

Distro Version

Ubuntu 24.04

Other Software

CUDA 12 as loaded by TensorFlow.
NVIDIA Studio Driver 595.79.

Repro Steps
$ python3 -m venv /tmp/repro/
$ . /tmp/repro/bin/activate
$ pip install 'tensorflow[and-cuda]'
$ LD_DEBUG=libs python3 -c 'import tensorflow as tf; print(tf.config.list_physical_devices("GPU"))' 2> ld_debug.log
Expected Behavior

TensorFlow detects the NVIDIA GPU.

Actual Behavior

It does not, presumably because of these errors:

       320:     calling init: /usr/lib/wsl/drivers/nv_dispsi.inf_amd64_fe1c50c690e00e9a/libcuda.so.1.1
       320:
       320:     /usr/lib/wsl/drivers/nv_dispsi.inf_amd64_fe1c50c690e00e9a/libcuda.so.1.1: error: symbol lookup error: undefined symbol: cuPvtCompilePtx (fatal)
       320:     /usr/lib/wsl/drivers/nv_dispsi.inf_amd64_fe1c50c690e00e9a/libcuda.so.1.1: error: symbol lookup error: undefined symbol: cuPvtBinaryFree (fatal)
       320:
       320:     calling init: /usr/lib/wsl/drivers/nv_dispsi.inf_amd64_fe1c50c690e00e9a/libnvdxgdmal.so.1
       320:
       320:     /usr/lib/wsl/lib/libdxcore.so: error: symbol lookup error: undefined symbol: D3DKMTOpenSyncObjectFromNtHandle (fatal)
       320:     /usr/lib/wsl/lib/libdxcore.so: error: symbol lookup error: undefined symbol: D3DKMTAcquireKeyedMutex (fatal)
       320:     /usr/lib/wsl/lib/libdxcore.so: error: symbol lookup error: undefined symbol: D3DKMTReleaseKeyedMutex (fatal)
       320:     /usr/lib/wsl/lib/libdxcore.so: error: symbol lookup error: undefined symbol: D3DKMTOpenKeyedMutex (fatal)
       320:     /usr/lib/wsl/lib/libdxcore.so: error: symbol lookup error: undefined symbol: D3DKMTRegisterTrimNotification (fatal)
       320:     /usr/lib/wsl/lib/libdxcore.so: error: symbol lookup error: undefined symbol: D3DKMTUnregisterTrimNotification (fatal)
       320:     /usr/lib/wsl/lib/libdxcore.so: error: symbol lookup error: undefined symbol: D3DKMTCreateNativeFence (fatal)
       320:     /usr/lib/wsl/lib/libdxcore.so: error: symbol lookup error: undefined symbol: D3DKMTOpenNativeFenceFromNtHandle (fatal)
Diagnostic Logs

ld_debug.log

WslLogs-2026-03-16_21-34-21.zip

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the issue with the listed Python virtual-environment and TensorFlow installation commands, then inspect ld_debug.log and the attached WslLogs archive. Focus on the reported CUDA and libdxcore undefined-symbol errors and compare behavior across the listed WSL, kernel, driver, and Ubuntu versions. Done means TensorFlow detects the NVIDIA GPU without those loader errors.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, tensorflow
Domain
operating-systems
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.