Hi,
I’ve recently done a PC build and successfully installed Fedora as the Linux distribution. My build has an Intel i5 12600k (with integrated GPU) and Nvidia GEforce RTX 3060. Both of which have drivers installed.
The PC will be primarily used for machine learning and AI tasks so I want to use the integrated GPU for all graphics/display related stuff and reserve the dedicated GPU for use by the machine learning/AI programs.
At the moment, when I go to ‘Settings → About’ the graphics sections states ‘NVIDIA GeForce RTX™ 3060 / NVIDIA GeForce RTX™ 3060’ which seems a bit strange considering I only have one 3060 and the iGPU isn’t mentioned.
On doing a ‘nvidia-smi’, it states that there are two processes being run:
| 0 N/A N/A 2037 G /usr/bin/gnome-shell 146MiB |
| 0 N/A N/A 2382 G /usr/bin/gnome-software 7MiB |
Maybe the about section only shows the dedicated GPU cause that’s the GPU that is running GNOME, I believe.
How can I restrict gnome to the integrated GPU, and reserve the dedicated GPU for AI/Machine learning tasks.
In response to 1. No, I don’t believe it is manually disabled in BIOS cause I’ve been trying different settings to make sure it wasn’t.
As for 2. Initially, when I first installed Fedora, it was the Integrated GPU showing up in settings → about. It was only after installing the nvidia driver that the integrated gpu seemed to be disabled. So yeah, you could be right.
It has been only connected to the motherboard, not directly to the GPU. Since I’ll only be using the dedicated GPU for Machine learning/AI work.
As for the last question, I’ll get back to you. I’m after locking myself out, messing around with settings, so am going to have to do a fresh install, haha
I’ve managed to get both my integrated and dedicated GPU showing in ‘settings → about’. I connected the monitor directly into motherboard and changed the BIOS to use integrated GPU by default with multiple monitors disabled.
Thanks everyone for your help.
I’m still getting the following though when I do a ‘nvidia-smi’.
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.85.05 Driver Version: 525.85.05 CUDA Version: 12.0 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA GeForce ... Off | 00000000:01:00.0 Off | N/A |
| 0% 32C P8 12W / 170W | 3MiB / 12288MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 2061 G /usr/bin/gnome-shell 2MiB |
+-----------------------------------------------------------------------------+
Ideally, I’d like to limit the dedicated GPU to machine learning tasks, but in the above you’ll see it’s involved in a gnome-shell process.
I don’t think your pc motherboard support optimus prime offloading. The only way I suppose is to use a switcher to disable your nvidia gpu when you want everything confined on your iGPU.
But is there a reason for doing so? You won’t benefit much from doing such settings.