I have the below error on WSL for Linux on a windows 11 system, has anyone seen this before? Please advise, thanks!
Thu May 26 10:03:50 2022
±----------------------------------------------------------------------------+
| NVIDIA-SMI 515.43.04 Driver Version: 472.56 CUDA Version: 11.4 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA GeForce … Off | 00000000:01:00.0 On | N/A |
| N/A 41C P8 13W / N/A | Function Not Found | ERR! Default |
| | | N/A |
±------------------------------±---------------------±---------------------+
±----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
Hello,
Your Nvidia-smi version and your Driver version seem quite off. It usually happens when you install the native components (either Native Nvidia-smi or Native Driver) on a WSL system rather than using the one pre-installed.
To double check you can run:
strace nvidia-smi 2>&1 | grep “libnvidia-ml”
and check where the library ‘libnvidia-ml’, which is the library used by nvidia-smi to query those information, is coming from.
- If you see it coming from “/usr/lib/wsl/lib/” or from “/usr/lib/wsl/drivers/” the you have the right driver and something else is going. In that case make sure nvidia-smi itself is located in ‘/usr/lib/wsl/lib’.
- If you see it coming from a different location then you likely have a Native Linux Driver installed on the top of your WSL distribution and it is getting picked up rather than the WSL driver. Simply un-install the Native driver and you should see the right result.
If you still see the issue and if the driver is still reporting 472. Then it might be your native display driver on Windows being too old since 472 was quite early in the CUDA on WSL development. Try updating your Display Driver on the Windows side (it will automatically update the WSL one at the same time).
Thanks !
Hi @rboissel Thanks for your reply and the information. I get nothing when I run that command when I just run strace nvidia-smi 2>&1 I see error as below
Hello, NVIDIA team
I would like to follow up on this question because I had a similar problem. I used WSL (Ubuntu 20.04) for linux on a windows 10 (GeForce GTX 1660 Super).
Here is the version of my NVIDIA:
NVIDIA-SMI 515.65.01 Driver Version: 516.94 CUDA Version: 11.7
When I typed nvida-smi in WSL, it showed the NVIDIA version but also said
No running processes found for the Processes
and when I typed strace nvidia-smi 2>&1 | grep “libnvidia-ml”, it said (please also checked the attached photo)
strace: Can’t stat ‘nvidia-smi2’: No such file or directory
I am not sure to what extent this affects Tensorflow but when I tested Tensorflow via the latest Docker image with command such as in the ipython environment
check_gpu = len(tf.config.list_physical_devices('GPU'))>0
it showed warning that
could not open file to read NUMA node: /sys/bus/pci/devices/0000:08:00.0/numa_node Your kernel may have been built without NUMA support.
Therefore I am really confused and would really appreciate some guidance to point out where the problem is. Thank you!
