Optimization of Stable Diffusion 1111 #11799
Replies: 2 comments
-
The latest Nvidia drivers do that automatically, it's slower because they spill over to normal RAM, but you won't get out of memory errors. Though currently the drivers have an issue where they start slowing down even when VRAM isn't completely full. So if you have 12GB of VRAM, it sometimes slows down at around 10GB of usage. If you can wait a couple days, there might be a new driver this week that may or may not address this issue, the 4060ti 16gb launches this week and they usually pair new card releases with a new driver. You can also just try --medvram or even --lowvram launch arguments. https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings |
Beta Was this translation helpful? Give feedback.
-
Are you generating images at that resolution without using upscale? If you use, for example, Tiled Vae you can generate images even at 8k resolution on 6Gb without memory errors. |
Beta Was this translation helpful? Give feedback.
-
Hello everyone, my name is Roberto
and recently I became interested in the generation of images through
the use of AI, and in particular with the Automatic 1111 distribution of Stable Diffusion.
I would like to ask you a question, I often find myself dealing with memory error problems when I try to create images on my Windows 10 PC with my Nvidia 3060Ti (8Gb Ram) graphics card.
The most I got is a resolution of 1184x1184.
But often, if I interrupt the calculation, perhaps because I realize that the image being calculated does not meet my expectations, and I interrupt the calculation, at the next recalculation I get a memory error. So I'm forced to restart the pc to be able to try again to generate a new one with the same prompt.
Is there a way to avoid this long and tedious procedure?
Often then I try to inpaint an already generated image (perhaps extended through stable diffusion), (sometimes even resizing an image) to run into the "cuda-out-of-memory" error again. Is there any way to avoid it?
Is it possible to use the PC Ram to increase the amount of memory available for image generation? or do I have to resign myself to waiting (a lot) before being able to buy a more powerful graphics card?
Thanks for any suggestions you can get me.
Roberto
P.S.
I find Stable Diffusion to be a spectacular piece of software, which I have found very useful for generating backgrounds and characters for my favorite RPGs (for campaigns with my friends). I'm curious to see what AI image generation will be able to do in a few years.
Beta Was this translation helpful? Give feedback.
All reactions