16 series graphics cards output error and can not output larger resolution problem

[baiwei666]

The output of a 16 series graphics card is a green picture, how to solve it? And 6g video memory cannot output pictures with a resolution of more than 512*512. How to set it to break through this limit?

[ack226]

You'll need to use the following options:

--precision full --no-half --optimized

I have a GTX 1660 Super and I was able to use --optimized-turbo instead of optimized, which works fine and made it run much faster, so you should try that and if it crashes revert back to the slower --optimized option.

I also added --extra-models-cpu to force the upscaling models to use CPU since my GPU doesn't have enough memory.

My full arguments line for the GTX 1660 Super (6GB) is:

--precision full --no-half --optimized-turbo --extra-models-cpu

With the above everything seems to work great!

[baiwei666]

python scripts/webui.py --precision full --no-half --optimized-turbo --extra-models-cpu
I've added that code and it outputs a normal image, but still can't generate a higher resolution image. Program prompts me “CUDA out of memory”

[ack226]

As far as I know, 512x512 is about as high as you'll be able to go with 6GB of video memory. Most people, even with better cards, generate 512x512 images and then upscale them afterwards.

[ignotusandri]

Hey :) I'm having similiar problems.
I'm running stable diffusion on a Laptop with a GTX1650 Ti with 4GiB of vram. When using optimized or optimized turbo I am able to run it, however I only get a green square image as a result. Following the advice here I've added the suggested code. However now when I het the generate button I immediately get a crash message due to a lack of CPU.

[mchaker]

4GB may be too small to run at full precision