Text-to-Image
stable-diffusion

GPU specs to run the model

#6
by wass-grass - opened

Hello, the title says it all. Which minimal system requirements needed to run this model ? Which GPU and How much VRAM approximately?

Many thanks

I think the minimum requirements are about 6900 MiB of GPU memory (see the fp16 branch). That's also used for this colab.

For a safe bet, the announcement blog originally said 10gb, now there is some twitter discussion around 5.1 gb, but not sure how to do that.

for weights version 1.4 with code from https://github.com/hlky/stable-diffusion/ and https://github.com/hlky/stable-diffusion-webui/ you can run with half precision and needs up to 8 Gb VRAM for 512x512 image size
with "webui.py --optimized" it run iterative and need only about 4 Gb VRAM, but works about 4 times slower than 8Gb VRAM version
also you can look into Stable Diffusion Reddit wiki to check for new CPU-only versions

Sign up or log in to comment