What is the oldest support card version?

I have an old card, about 800 series, it`s on a laptop, did it be supported?

Could be either Fermi or Kepler.

go to advanced tab / CUDA
check the compute capability

None of these mobile GPUs is supported in CUDA on WSL drivers though.

Hi StefanG3D,

None of these mobile GPUs is supported in CUDA on WSL drivers though.

Where did you find this information?

CUDA on WSL :: CUDA Toolkit Documentation says:
“Note: CUDA on WSL 2 is enabled on GPUs starting with the Kepler architecture; however, we recommend running CUDA on WSL 2 on Turing or newer architectures.”

So I would have thought that e.g. my Gigabyte Laptop’s GeForce GTX980M, having a Maxwell Architecture (i.e. newer than Kepler) should be supported?

Also, do you know wether there are any limitations w.r.t. CUDA on WSL2 for external full-size GPUs which however are mounted in a separate box connected through Thunderbolt3 or a proprietary PCIe connector?


Fermi support (SM 2.0) has been physically removed in R396+ drivers, as documented in CUDA 9.2 release notes

Regarding Mobile Kepler review Support Plan for Kepler-series GeForce GPUs for notebooks .
As you can see in ListDevices.txt inside the driver package mobile Keplers are missing.

“Binary compatibility within one GPU generation can be guaranteed under certain conditions because they share the basic instruction set. This is the case between two GPU versions that do not show functional differences at all (for instance when one version is a scaled down version of the other)”
Source: CUDA Compiler Driver NVCC Reference Guide

Developers should know how to help themselves…

Maxwell mobile is supported in R470 drivers, actually i installed it on a 840m (Maxwell 1st gen)

For EGPU issues visit https://egpu.io/

The upthread question refers to a mobile Maxwell GPU. I can confirm that WSL2 works on those.
After several failed trials I, too, had developed doubts whether mobile GPUs are fully supported though.
The culprit turned out to be the Windows build I used, 21354 broke the WSL2 support.
On the newest build (21359) the CUDA examples work. Although nvidia-smi still can’t connect to the NVIDIA driver.