ZLUDA on dev branch offloads to cpu instead of gpu #2869

dragitz · 2024-02-16T14:19:28Z

dragitz
Feb 16, 2024

Hello, I know it's not the issue tab but I've read that if there's an issue related to the dev branch, it is better to either discuss it here or on discord. Unfortunately the invite link has expired so I'm making the post here, apologies in advance if it's not the right place :)

First of all, thank you very much for your work and the detailed guide provided here https://github.com/vladmandic/automatic/wiki/ZLUDA

I follwed every step and the last step took around 14 minutes to compile.

After the compilation ended, in the console I saw an incredible boost in it/s

I've checked if ZLUDA was properly installed and it seems to recognize the gpu without any issue

But after generating the first image the iterations were identical as if I was using --use-directml combined with my gpu (~2 it/s)

I've checked every setting containing the words such as "cpu, gpu or offload" in the system tab but none of them were the issue.
I did notice that in the console it said low_cpu_mem_usage: True so I took the liberty of editing this line to ensure it wasn't the issue, but unfortunately I had no success (this is why it is set to False in the logs)

Logs:
sdnext.log

Version Platform Description

Dev version: 3c95267
OS: Windows 10.0.19045 Build 19045
Browser: Firefox 122.0.1
GPU: 6700 xt (gfx1031)
CPU: 5600x

torch. __version__ 2.2.0+cu118
Extensions enabled: ['Lora', 'sd-extension-chainner', 'sd-extension-system-info', 'sd-webui-agent-scheduler', 'sd-webui-controlnet', 'stable-diffusion-webui-images-browser', 'stable-diffusion-webui-rembg']

SD Model based on 1.5

I've also restarted my computer twice just to be sure I've done everything correctly, and my installation is the default one with the exception of the above mentioned edit.
I'm sure something is missing on my side, and I'm struggling to find it, what could the issue be? Does anyone have a similar experience?