Upgrading to Nvidia A100 does not improve it/s

[danileon97]

Recently I have changed from a NVIDIA V100 16GB to an NVIDIA A100 80GB PCIe GPU. I have configured everything as I did with the previous GPU. I am running the sd-xl-base-1.0 model.

For both GPUs I can see how the GPU usage when running inference does not go above 10GB. This results in 2-3 it/s in both GPUs.

How can I take profit of the whole 80GBs? Do I have to configure something? I am sugin the following command to run the webui:

./webui.sh --listen --share --api --api-log --enable-insecure-extension-access

Thanks!

[playlogitech]

more ram doesn't mean faster it;'s
you just got more space to use without stupid tricks like tiled vae,etc

[YuxuanWang0822]

@danileon97
Hi, I also change V100 to NVIDIA A100 but the generation speed becomes extremely slow, expecially during sampling.
Do you have the same issue?

[catclaw]

I'm having the same issue with Fooocus...
I upgraded to PNY Nvidia Tesla A100 80 GB, but it's just as slow as RTX 3060 12 GB?