This thread is permanently archived
Running stable diffusion in the backgroubd
| firstly: what sort of cursed os are you running for the kernel to not be able to schedule vram properly?
second: sd's vram usage depends 100% on the resolution of the images you are generating. up to 1024x1024 *should* be fine on 8 gigs? but i would advise you to do a batch of 20 on a given resolution and monitor terminal to see how much is being allocated each time, get averages, and get your hard data that way.
| remember you can use the split attention xformer to help drop vram at the cost of generation speed too.
| I tried it months ago on windows with AMD GPU with 4GB of VRAM (no, I did not expect it to go well, Just curious about how long it will takes to generate one image)
No idea what setting I used, no idea on the image resolution or setting (probably just default stuff), I just fumbled and got it working. PC unusable, so I never looked at it again.
I am planning to buy a new PC 3 months from now. Still wondering if I should go with 12GB 4070 or the upcoming 16GB 4060.
This thread is permanently archived
| Most of the time it consumes 8GB of VRAM, right? If I have a GPU with 10GB Or 12GB of VRAM, can I run stable diffusion in the background without crippling my PC?
I've tried before using a GPU less than 8GB of VRAM and it turns other programs unusable.
I don't mind if it makes the generations slower, I just want to do other tasks like watching a video or play easy to run RPG maker games while it generates image. Can anyone who own a GPU with more than 8GB VRAM confirm this?