🐢 Slow performance? / GPU utilization🧾 DescriptionI’m using a 2x batch image upscaling feature with the Remacri model on 1920×1080 JPG images. The process feels slower than expected. it takes about 65 seconds per image. 🖥️ System setup
⚙️ Configuration
📊 Observed behaviorEven when running 3 instances in parallel:
System usage during parallel run:
❓ QuestionIs this expected behavior or a limitation of the program/model? Is there any way to better utilize my hardware to improve processing speed? Any insights would be appreciated. 📝 NoteMy English writing skills are not very strong, so I used ChatGPT to help structure this message. |
Replies: 1 comment 2 replies
|
Hi @mia64! I'm Dosu and I'm helping the Upscayl team. With an RTX 3060 12GB, ~65 seconds per 1080p image at 2x with Remacri does seem slower than expected, and the very low GPU usage (~10%) strongly suggests Upscayl may be falling back to software/CPU rendering instead of properly using your GPU. First, verify your GPU is actually being used:
Other speed optimizations:
The ~10% GPU utilization is the biggest red flag here. Once your GPU is properly utilized, you should see much faster processing times on an RTX 3060 with 12GB VRAM. To reply, just mention @dosu. Docs are dead. Just use Dosu. |
Thanks for the suggestions.
I followed your guide and got good results so far
What I did:
Set Upscayl to "High performance"
already set
Enable hardware-accelerated GPU scheduling
I'm using Windows 10, but I checked online and found how to enable it
Set GPU ID to 0
Already set.
Check for conflicting Vulkan software
To my knowledge, there aren't any programs running in the background except Firefox.
Clean reinstall NVIDIA drivers using DDU
Done. I followed this YouTube guide: https://www.youtube.com/watch?v=F8OLhUAPDq
Increase tile size
This seems to have made the biggest difference.
However, tile sizes 620 and 640 produce the following error: