Dual GPU for two different tasks

[Ido_sch]

Hi, I am trying to find information about multi GPU system where each GPU is dedicated to other tasks.

 

I currently have RTX 3060ti in my system and plan to upgrade to the 50 series soon.

I want to keep the my current GPU and dedicate it strictly for machine learning.

 

I have the gigabyte z690 gaming mother board, so I know I have additional slot for GPU (PCI 3.0 x4) and since I don’t want the secondary GPU to do anything but machine learning and research purposes I am okay with it.

everything find talks about combining the two GPUs for the same task.

[Blue4130]

1 minute ago, Ido_sch said:

Hi, I am trying to find information about multi GPU system where each GPU is dedicated to other tasks.

 

I currently have RTX 3060ti in my system and plan to upgrade to the 50 series soon.

I want to keep the my current GPU and dedicate it strictly for machine learning.

 

I have the gigabyte z690 gaming mother board, so I know I have additional slot for GPU (PCI 3.0 x4) and since I don’t want the secondary GPU to do anything but machine learning and research purposes I am okay with it.

everything find talks about combining the two GPUs for the same task.

Do you plan on machine learning while you are gaming?

[Ido_sch]

16 minutes ago, Blue4130 said:

Do you plan on machine learning while you are gaming?

I might, which is why I want complete separation (I also made sure to have enough RAM)

[manikyath]

if your software can talk to a GPU directly it's just a matter of having the GPU present, and telling the software to use that one.

 

if your software relies on the OS to direct it's calls to "the GPU", you're probably SOL.

 

in essence a GPU is just a device in the system, it's up to the software to support directing calls to a specific piece of hardware. in this sense it's not too dissimilar from how multiple storage drives work: it's up to the software to support not having your stuff saved in "my documents".

[Ido_sch]

13 minutes ago, manikyath said:

if your software can talk to a GPU directly it's just a matter of having the GPU present, and telling the software to use that one.

 

if your software relies on the OS to direct it's calls to "the GPU", you're probably SOL.

 

in essence a GPU is just a device in the system, it's up to the software to support directing calls to a specific piece of hardware. in this sense it's not too dissimilar from how multiple storage drives work: it's up to the software to support not having your stuff saved in "my documents".

So basically if I write a python script to do some model training, if there is a GPU present I can tell it to use it instead of my cpu?

[igormp]

On 4/26/2025 at 8:08 AM, Ido_sch said:

, if there is a GPU present I can tell it to use it instead of my cpu?

Yes, you can just say it to use "cuda:0" or "cuda:1" as your device.