Selecting second GPU

[GrewsomeC]

Good evening,

I just installed a second GPU into my machine, a RTX 3060 with 12gb VRAM. This is a significant improvement over my 2070S with only 8gb VRAM that it has to share with Windows.

But I can't figure out how to actually select it. There's nothing in settings.

I have tried every solution I could find by googling, such as:

• set CUDA_VISIBLE_DEVICES=1 (This does nothing -- webui still uses GPU 0, my 2070)
• changing devices.py line 15 from "cuda" to "cuda:0" or "cuda:1" (This is impossible -- devices.py line 15 is blank)
• set COMMANDLINE_ARGS= --device-id=1 (This does nothing -- webui still uses GPU 0, my 2070)
• Reinstalling pytorch (This does nothing -- webui still uses GPU 0, my 2070)

Notably, if I try to do something like --device-id=2, the software (correctly) throws an error. This means that it DOES recognize there is a second GPU and that I have selected it, but chooses to ignore it for some reason and instead swap, with no error or reason, back to GPU0.

So, any assistance would be lovely! All the previous solutions simply don't function. I appreciate it.

[chrme]

edit webui-user.bat like this:

set CUDA_VISIBLE_DEVICES=0,1
set COMMANDLINE_ARGS=--device-id 1

[GrewsomeC]

Solution found. For some reason, webui sees the video cards as the other way around. Despite my 2070 being GPU 0 and my 3060 being GPU 1 in Windows, using --device-id=0 uses GPU1, while --device-id=1 uses GPU0.

No idea why, but that was the solution.

[scooller]

i try the same but always take the first GPU (Intel HD) not Nvidia dedicate in my laptop

[yun15634]

SDXL (1024x1024) and SDXL Training (1024x1024 Unet only) Works

GPU1=GT635 (Connected DL-DVI Monitor)
GPU2=RTX3060 (Not Connect Monitor, VRAM Efficiency, for SD only)
OS=Windows11
Driver=CUDA 11.4
1111=python: 3.10.11, torch: 2.0.1+cu118, xformers: N/A, gradio: 3.41.2, checkpoint: [1449e5b0b9]
(1449e5b0b9de87b0f414c5f29cb11ce3b3dc61fa2b320e784c9441720bf7b766)

Start - Settings - Game - Graphics Settings -> GPU Affinity - Select to Secondary GPU for Python.exe
C:\users\UserName\Appdata\Local\Programs\Python\Python310\Python.exe
C:\stable-diffusion-webui\venv\Scripts\python.exe
C:\LoRA_Easy_Training_Scripts\venv\Scripts\python.exe

nVidia Control Panel CUDA GPU Affinity, Select Secondary GPU
C:\users\UserName\Appdata\Local\Programs\Python\Python310\Python.exe
C:\stable-diffusion-webui\venv\Scripts\python.exe
C:\LoRA_Easy_Training_Scripts\venv\Scripts\python.exe

[Seedmanc]

Is there a way to benefit from having two dedicated GPUs in WebUI? I'm considering a similar upgrade to OP's, but I want to get more for my buck than just freeing up space occupied by Windows, I already have AMD video core for that.