New CUDA on WSL2 WIP driver 470.76 is now available for download!

Accelerated Computing CUDA CUDA on Windows Subsystem for Linux
1

We know you’ve been waiting for this update. We are here with a brand new WSL2 WIP driver 470.76.
Please download the driver from the download area

Refer to the release notes in the user guide for details on the changes that went into this release.

As always please let us know your feedback

Please register for the NVIDIA Developer Program and the Microsoft Windows Insider Program , to access the driver installers and documentation from the Downloads area on our CUDA on WSL webpage.

Our user guide and blog links below, contain valuable information to aid you to learn how CUDA works with WSL, including how to get started with running applications, and deep learning containers.

We encourage all of our developers to use our Forum for sharing their experience with the larger WSL community.

3 Likes
2

Thanks a lot!
No more nvidia-docker2 driver error
no more need for NVIDIA_DISABLE_REQUIRE=1
and nvidia-smi finally works (except GPU fan)!

$ nvidia-smi
Sat Jun  5 00:13:28 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.28       Driver Version: 470.76       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 N/A |                  N/A |
|ERR!    0C    P8    N/A /  N/A |    693MiB /  1024MiB |     N/A      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+
3

Thank you!
nvidia-smi works fluently now.

$ nvidia-smi
Sat Jun  5 06:58:28 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.28       Driver Version: 470.76       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 Off |                  N/A |
| N/A    0C    P8    N/A /  N/A |     64MiB /  2048MiB |    ERR!      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+
4

With this new driver version, its possible to play games and also able to use cuda on WSL2. Thank you!!

5

Can anyone advise on using tensorflow-gpu (any version) with this driver? The update to CUDA version 11.4 seems to have broken my tensorflow setup in WSL2 (‘cannot find libcusolver.so.11’) and it appears that no versions of tensorflow are yet compatible with CUDA 11.4 - 從原始碼開始建構  |  TensorFlow. I thought that specifically installing compatible older version of cudatoolkit and cudnn in the conda environment would fix it, but it hasn’t - what am I missing?

6

@cwkirkwood90 Did you try with nvidia-docker? I had no problems with docker run -u $(id -u):$(id -g) -it --rm --runtime=nvidia tensorflow/tensorflow:latest-gpu bash with driver 470.76.

The version of TensorFlow is 2.5.0 and these are the cuda libraries in Ubuntu 18.04 inside the container:

# apt search cuda | grep -i installed

cuda-command-line-tools-11-2/unknown,unknown,now 11.2.2-1 amd64 [installed]
cuda-compat-11-2/unknown,now 460.73.01-1 amd64 [installed]
cuda-cudart-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed]
cuda-cudart-dev-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-cuobjdump-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-cupti-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-cupti-dev-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-driver-dev-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-gdb-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-memcheck-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-nvcc-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-nvdisasm-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-nvprof-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-nvrtc-11-1/unknown,unknown,now 11.1.105-1 amd64 [installed,automatic]
cuda-nvrtc-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed]
cuda-nvtx-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
cuda-sanitizer-11-2/unknown,unknown,now 11.2.152-1 amd64 [installed,automatic]
libcusolver-11-2/unknown,unknown,now 11.1.0.152-1 amd64 [installed]
7

Thanks! Will try this. Does this update allow full use of GPU memory?
For example, if I have 24GB of GPU memory, I want to use all GPU memory.