NVIDIA Adds Official Support for Virtualizing GeForce GPUs


Recommended Posts

Big News from NVIDIA

Just a few hours ago, NVIDIA added an article to its support knowledge base regarding GPU passthrough support for Windows VMs. While we've supported this functionality for some time, it was done without official support from the vendor themselves.

This move by NVIDIA to announce official support for this feature is a huge step in the right direction for all of our VM pass through users. This should also help instill confidence in users that wish to pass through these GPUs to virtual machines without the worry that a future driver update would break this functionality.

 

Let us know what you think about this announcement here!

 

  • Like 8
  • Thanks 3
Link to comment
7 hours ago, tjb_altf4 said:

Hopefully this opens up the possibility of SR-IOV at a later date

 

I don't think so because they also said:

Quote

Do you need to have more than one GPU installed or can you leverage the same GPU being used by the host OS for virtualization?
One GPU is required for the Linux host OS and one GPU is required for the Windows virtual machine.

 

and also:

Quote

What class of virtualization is supported on GeForce GPUs?
GeForce GPU passthrough supports 1 virtual machine.  SR-IOV is not supported on GeForce.  If you want to enable multiple virtual machines to have direct access to a single GPU or want the GPU to be able to assign virtual functions to multiple virtual machines, you will need to use NVIDIA Tesla, Quadro, or RTX enterprise GPUs.

 

I think they will not support the feature anywhere in the near future... But this is the first step in the right direction sort of. :D

  • Like 2
Link to comment

Do I still have to pass the vBIOS to the VM?

 

I've been having issues with GPU pass-through and a 1050 Ti where restarting/shutting down a VM will kill the entire unRAID server requiring a restart and parity rebuild.  With 14TB drives that very frustrating.

 

I want to try with this new feature to see what happens as I'd love to have a stable VM with GPU pass-through. 

 

 

ETA: It appears you can leave off the vBIOS if you switch to VNC, start the VM, stop the VM, and then switch to GPU w/o specifying the vBIOS ROM.  I just got a black screen if I didn't start the VM in VNC mode first.  Now to see if my VM issue goes away.

Edited by jlficken
Link to comment
16 hours ago, jlficken said:

Do I still have to pass the vBIOS to the VM?

 

I've been having issues with GPU pass-through and a 1050 Ti where restarting/shutting down a VM will kill the entire unRAID server requiring a restart and parity rebuild.  With 14TB drives that very frustrating.

 

I want to try with this new feature to see what happens as I'd love to have a stable VM with GPU pass-through. 

 

 

ETA: It appears you can leave off the vBIOS if you switch to VNC, start the VM, stop the VM, and then switch to GPU w/o specifying the vBIOS ROM.  I just got a black screen if I didn't start the VM in VNC mode first.  Now to see if my VM issue goes away.

I think we always need th vBIOS, just, it's not necessary to modify it with an HEX editor to use it. But I'm not sure of what I say.

Link to comment

It seems like a big news, but I just wonder what exactly it changes since we could already use the passthrough before, like show in some many videos and like I did for a GTX 1060 and a 1070.

Is it a different way to setup now?
Or were there some older cards that couldn't be passthough no matter what?

Link to comment
On 4/1/2021 at 5:44 PM, dada051 said:

Why "Now if only AMD would make the same gesture... 😉" ? I didn't know that AMD locked the use of pass-through

AMD fixed the reset bug on latest gen of GPUs, so no idea, what the author is talking about.

Link to comment
On 4/1/2021 at 11:44 PM, dada051 said:

Why "Now if only AMD would make the same gesture... 😉" ? I didn't know that AMD locked the use of pass-through

They didn't lock it, they just refused to fix a bug (hardware reset) that made pass-through unusable.

As soon as the community started to build working fixes, AMD's next generation didn't have these issues, but the fixes weren't backported to previous gen.

 

I should also note these hardware reset problems have never appeared in AMD's professional GPU lineup (Quadro competitors), funny that.

 

  • Like 1
Link to comment
  • 1 month later...
8 hours ago, Kosmos said:

This would actually be the thing I'm looking for.

I would like to assign one gpu to multiple VMs.

As @tjb_altf4 said, this thread has nothing to do with assigning one GPU to multiple VM's.

 

8 hours ago, Kosmos said:

Can we somehow make this work? Maybe as an Unraid plugin?

rel: https://github.com/DualCoder/vgpu_unlock

This violates the Nvidia EULA, I think you know what that means...

Also keep in mind you have to buy licenses from Nvidia for every VM that you want to run a vGPU on.

  • Thanks 1
Link to comment
  • 1 month later...
  • 3 weeks later...
  • 3 weeks later...
On 7/7/2021 at 11:54 PM, jbquintal said:

Any ideas as to when we could potentially see this natively supported in unRAID?

I don't think you will see this natively on unRAID since this violates the EULA from Nvidia.

Link to comment
On 7/26/2021 at 12:29 AM, ich777 said:

I don't think you will see this natively on unRAID since this violates the EULA from Nvidia.

I understand the hack violates the EULA, but I'm curious about how Hyper-V can now support GPU-P which accomplishes the same thing (on both Nvidia and AMD) without creating issues with licensing.

We might get lucky and similar changes get adopted in KVM.

 

Another video from Jeff for context 😛

 

 

  • Like 1
Link to comment
1 hour ago, tjb_altf4 said:

We might get lucky and similar changes get adopted in KVM.

I already saw that but I think it's mainly based on that: Click

 

Also keep in mind that the author of this mentioned:

Quote

Apparently, it works with Nvidia drivers that came out before the VM passthrough enablement.

So I don't think Nvidia will allow this on consumer cards and will update their EULA accordingly.

 

1 hour ago, tjb_altf4 said:

I understand the hack violates the EULA

I think it will always violate the EULA and if a way is found that doesn't violates the EULA, Nvidia will always update the EULA that it will violate the EULA (that is a confusing sentence but I hope it makes sense... :D ).

 

 

Keep in mind everything that Jeff did here will basically work on unRAID with a few differences, but that will always violate the EULA. :/

Link to comment
  • 4 weeks later...

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.