You are not logged in.

#1 2024-08-15 14:19:18

nv3r
Member
Registered: 2024-08-15
Posts: 2

Multi-GPU usage after passthrough

I am curious if the following is possible. Currently, I have my Nvidia GPU setup for pass-through to my windows KVM. I'm using an AMD GPU within my main Linux system. I want to know if I can use my isolated Nvidia GPU for tasks such as Stable Diffusion on the main machine when it is isolated or if I will need to have both GPU's setup for the main system ~ and if I did have a dual GPU setup, would this then prevent pass-through to the KVM?

If needed, here are the exact gpus on my system:

$ lshw -c display                                                                   

  *-display                 
       description: VGA compatible controller
       product: GA102 [GeForce RTX 3080 Ti]
       vendor: NVIDIA Corporation
       physical id: 0
       bus info: pci@0000:01:00.0
       version: a1
       width: 64 bits
       clock: 33MHz
       capabilities: vga_controller cap_list rom
       configuration: driver=vfio-pci latency=0

  *-display
       description: VGA compatible controller
       product: Navi 31 [Radeon RX 7900 XT/7900 XTX/7900M]
       vendor: Advanced Micro Devices, Inc. [AMD/ATI]
       physical id: 0
       bus info: pci@0000:07:00.0
       logical name: /dev/fb0
       version: ce
       width: 64 bits
       clock: 33MHz
       capabilities: vga_controller bus_master cap_list rom fb
       configuration: depth=32 driver=amdgpu latency=0 resolution=5120,1440

  *-display
       description: VGA compatible controller
       product: Raphael
       vendor: Advanced Micro Devices, Inc. [AMD/ATI]
       physical id: 0
       bus info: pci@0000:15:00.0
       version: c1
       width: 64 bits
       clock: 33MHz
       capabilities: vga_controller bus_master cap_list
       configuration: driver=amdgpu latency=0



Overall I want to know if I can use my Nvidia GPU for tasks when my kvm is not running.

Last edited by nv3r (2024-08-15 14:24:39)

Offline

#2 2024-08-15 15:46:52

cryptearth
Member
Registered: 2024-02-03
Posts: 2,104

Re: Multi-GPU usage after passthrough

well - as long as the vfio driver is used for the gpu it will only provide pass-thru capabilities - as you need any gpu driver loaded for it to be recognized as gpu - otherwise it's just any non-gpu-device
I'm not sure about hot-swapping used driver - but I've seen scripts so it seems possible

Offline

#3 2024-08-16 13:09:36

nv3r
Member
Registered: 2024-08-15
Posts: 2

Re: Multi-GPU usage after passthrough

cryptearth wrote:

well - as long as the vfio driver is used for the gpu it will only provide pass-thru capabilities - as you need any gpu driver loaded for it to be recognized as gpu - otherwise it's just any non-gpu-device
I'm not sure about hot-swapping used driver - but I've seen scripts so it seems possible

Got it, I seemed to have gotten my desired functionality by following the hybrid graphics walkthrough. I can still pass-through to the KVM, but can prime-run when the KVM isn't running.

Offline

#4 2024-08-16 23:29:48

cloverskull
Member
Registered: 2018-09-30
Posts: 277

Re: Multi-GPU usage after passthrough

Hey @nv3r, can you speak more to how you got this to work using prime-run with vfio?

As far as I can gather with my hybrid intel/nvidia setup, using the vfio driver for my nvidia card means that it can only pass through to the kvm and otherwise I'm unable to use it (i.e., with prime-run). Can you share your kernel command line and any other configuration steps that you did?

Thanks!

Offline

Board footer

Powered by FluxBB