Skip to content
This repository has been archived by the owner on Jan 22, 2024. It is now read-only.

NVIDIA docker on windows? #665

Closed
redBirdTx opened this issue Mar 13, 2018 · 19 comments
Closed

NVIDIA docker on windows? #665

redBirdTx opened this issue Mar 13, 2018 · 19 comments

Comments

@redBirdTx
Copy link

Hi,

Not really any issues but a question on OS. From past thread I read that NVIDIA docker is not for windows. Is it still the case now?

Thanks

best regards

@3XX0
Copy link
Member

3XX0 commented Mar 13, 2018

Yes see https://github.com/NVIDIA/nvidia-docker/wiki/Frequently-Asked-Questions#is-microsoft-windows-supported

@3XX0 3XX0 closed this as completed Mar 13, 2018
@npagare
Copy link

npagare commented Jul 4, 2018

Hi @3XX0, What's the reason to not to support Windows? Many of us have Windows laptop with dedicated NVIDIA GPUs such as GTX 1060. Having your support will be very valuable.

Please reconsider in supporting windows as well instead of routing us thru work arounds.

Thanks,

@chenzhekl
Copy link

@IoTFier Because there is no GPU passthrough in Hyper-V on Windows 10 Pro, I guess.

@dims12
Copy link

dims12 commented Dec 5, 2018

Incredible.

@lanwatch
Copy link

It would be useful still to have a devel image.

@llucid-97
Copy link

With process isolation being released for windows containers on windows 10, are there any plans to change this, and have GPU-enabled windows containers?

@timotei
Copy link

timotei commented Apr 27, 2019

@ihexx I think the issue here is not necessarily the process isolation, but rather GPU passthrough, since you'd need to be able to share the GPU somehow to the docker container. Also, it needs that 'nvidia' docker runtime, but haven't looked into details.

Also, there are two docker variants on Windows: Docker via HyperV Linux VM (where you'd 100% need GPU passthrough) and natively. For the native part I think we need the proper driver to be compiled for windows - but I can't see/find the source code for that to even try :(

PS: some possibly related topic on the internets w/r/t the GPU Passthrough for the Hyper-V variant of docker. Haven't managed to try them myself yet though.

https://social.technet.microsoft.com/Forums/en-US/96a52f97-dc8c-48ae-8a99-2bfdbc05c526/gpu-pass-through-on-windows-10-hyperv?forum=win10itprovirt
https://www.reddit.com/r/HyperV/comments/8jwt5c/windows_10_hyperv_dda/
https://gridforums.nvidia.com/default/topic/8989/general-discussion/list-of-gpus-that-support-dda-in-server-2016-for-gpu-pass-through-in-hyper-v-vms/

@MarcoVdE
Copy link

MarcoVdE commented Dec 2, 2019

Hi,

would this have any impact: https://docs.microsoft.com/en-us/virtualization/windowscontainers/deploy-containers/gpu-acceleration

Though it requires that it must be a windows container and not a Linux one. According to: #429 there is no GPU access in WSLv2, so can't attempt it with that either.

Dual booting or using a remote machine seem to be the only options currently for Windows users if wanting to use a docker image to maintain environment and do resource allocation.

@f1am3d
Copy link

f1am3d commented Mar 29, 2020

Yes see https://github.com/NVIDIA/nvidia-docker/wiki/Frequently-Asked-Questions#is-microsoft-windows-supported

99% nvida users on Windows

No, we do not support Microsoft Windows

image

What is the actual nonsense, nvidia?

@maximveksler
Copy link

I prefer nvidia to focus narrow and ship the best GPU support on linux then wide and attempt to hit Linux & Windows but releasing slower or buggier.

@rarvikar
Copy link

Can you please explain what do you exactly mean when you say "GPU passthrough is not available on Hyper-V" in Windows ?

"Passthrough mode", supported by Microsoft Hyper-V (and also Citrix XenServer and VMware ESX) simply hands off the entire GPU to the hypervisor for the benefit of one VM. However, no carving or segmentation is possible (which is what NVidia's vGPU feature does).

So considering that 90% of your customer base uses Windows are there any plans of adding support albeit one that conforms to Microsoft's "Passthrough Mode" to hand-off the entire GPU to the Hyper-V VM that hosts Docker containers ?
TIA !

@Pit-Storm
Copy link

I would although appriciate this feature for windows. But I know, that nearly 100% of servers (which this in most cases ist used for) are running linux. So this will not happen in near future...

@gedoensmax
Copy link

@Pit-Storm unexpected things happening ;) https://devblogs.microsoft.com/directx/directx-heart-linux/

@npagare
Copy link

npagare commented May 19, 2020

@Pit-Storm unexpected things happening ;) https://devblogs.microsoft.com/directx/directx-heart-linux/

@gedoensmax - yes - I am not sure when that change will hit Surface Book 2 with dGPU ?

@NicTanghe
Copy link

Ok but how does this help us ?
with us i mean users that are forced to use windows because of reasons and things.

@ShuyaoGu
Copy link

Would a Linux Virtual Machine inside Windows 10 make NVIDIA docker working?

@nvjmayo
Copy link
Contributor

nvjmayo commented Jun 17, 2020

see also https://developer.nvidia.com/cuda/wsl

@78Alpha
Copy link

78Alpha commented Jul 13, 2020

Was currently installing this and the runtime on WSL 2, it worked out up until I tried to download the Nvidia docker, GitHub apparently had a server issue and a good 20%? Of repositories couldn't be accessed, so I couldn't download the other half... However, docker and the runtime installed fine. The driver works and all. So it's possible, in every way, it's just simply a choice of support. Either by Nvidia or the community.

@PhearZero
Copy link

For anyone still digging this helps a lot! It's apparently in the pipeline and you need the latest Windows builds, Just verified TensorFlow in docker. Cheers!

https://docs.nvidia.com/cuda/wsl-user-guide/index.html#installing-wip

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests