Tiny little helper: we created a small tool to select the GPU to launch Automatic1111 with (Windows only) #11663
Replies: 8 comments 7 replies
-
very interesting. I thought you can choose it by writing something in the cmd-arg lines? Just installed it, thanks for sharing! Will use my second 3090 for more vram! |
Beta Was this translation helpful? Give feedback.
-
@48design hey! welcome back ! |
Beta Was this translation helpful? Give feedback.
-
You can mess around with the cmd arguments but using this tool you can just decide which GPU you like to use each time you run it so it may be a bit more useful. It's not restricted to Automatic1111 but can be used for every application you like. |
Beta Was this translation helpful? Give feedback.
-
@oliverban Does it work for you? |
Beta Was this translation helpful? Give feedback.
-
It does work just fine Nice with all that extra vram (about 2-3gb) and being able to do multiple things on the GPUs as well ! :) Now the problem is my other card is sort of starved for air so it gets hot as f*ck :P |
Beta Was this translation helpful? Give feedback.
-
The benefit in terms of vram sounds great! ;) |
Beta Was this translation helpful? Give feedback.
-
Unfortunately the tool does not work for me, and I tried several application executables using the context menu besides auto111. I select the 2060 card, but the 3060 is always used. Was really looking forward to it, too, for a couple of uses. As mentioned earlier though, just adding --device-id=1 to COMMANDLINE_ARGS works fine at least for auto111. Maybe something with how I have my dual monitors set up is just incompatible? |
Beta Was this translation helpful? Give feedback.
-
I updated the download and now this tool should work for paths with spaces, too. Have fun! |
Beta Was this translation helpful? Give feedback.
-
Since I have two old graphic cards (Nvidia GTX 980 Ti) and because Automatic1111/Stable Diffusion only use one GPU at a time I wrote a small batch file that adds a "GPU selector" to the context menu of Windows Explorer. This enables me to run Automatic1111 on both GPUs in parallel and so it doubles the speed as you can generate images using the same (or a different prompt) in each instance of Automatic1111.
I also tested creating a LoRA (using Kohya_ss) on one GPU and generating images on the other without problems!
One additional benefit: the second GPU has more VRAM available because it does not have additional OS related data in it.
You can download the tool here: https://www.48design.com/en/news/2023/05/16/ai-and-ml-arrived-at-everyday/
(scroll down to the end of the page)
I hope this helps others as well!
Beta Was this translation helpful? Give feedback.
All reactions