cancel
Turn on suggestions
Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for 
Search instead for 
Did you mean: 

AI Discussions

MrBP
Journeyman III
3 weeks ago

AMUSE 3.0.7. Kernel error

Hi,

After clear installation Amuse 3.0.7. downloading model StableDiffusion XL Turbo on my ThinkPad with AMD Ryzen AI 7 PRO 360 w/Radeon 880M (32GB) and after few successfully  generated pictures each next try for generated picture ended with Kernel error.

 

MrBP_0-1747818837974.png

Please help.

 

Thank you.

 

Petr

Labels
0 Likes
1 Solution
piper977
Journeyman III
3 weeks ago

 

Update the drivers for your GPU.
Visit AMD’s official site and grab the latest AMD AI assistant drivers for the Ryzen PRO 360 / Radeon RX 880M. Remember to check that Adrenalin Software is up to date as well.

Amuse lets you use DirectML Backend.
Go to Settings mode in Open Amuse and change the inference backend to DirectML (recommended for AMD GPUs). After you’ve made the change, restart Amuse.

Settings for Lower Gen.
Set the image size to 512x512, keep inference to only 20-25 iterations and experiment with Euler A or DPM++ 2M as the scheduler.

Every time you start a new generation, clean your VRAM.
Turn on memory cleanup or force the app to restart between any new generation to solve the issue.

Look for any updates from Amuse.
Be sure to use the newest version of Amuse, because updates usually take care of kernel and compatibility problems.

View solution in original post

2 Replies
piper977
Journeyman III
3 weeks ago

 

Update the drivers for your GPU.
Visit AMD’s official site and grab the latest AMD AI assistant drivers for the Ryzen PRO 360 / Radeon RX 880M. Remember to check that Adrenalin Software is up to date as well.

Amuse lets you use DirectML Backend.
Go to Settings mode in Open Amuse and change the inference backend to DirectML (recommended for AMD GPUs). After you’ve made the change, restart Amuse.

Settings for Lower Gen.
Set the image size to 512x512, keep inference to only 20-25 iterations and experiment with Euler A or DPM++ 2M as the scheduler.

Every time you start a new generation, clean your VRAM.
Turn on memory cleanup or force the app to restart between any new generation to solve the issue.

Look for any updates from Amuse.
Be sure to use the newest version of Amuse, because updates usually take care of kernel and compatibility problems.

AndewCarsten
Adept I
3 weeks ago

Good tips overall, especially the part about switching to DirectML for AMD users. I’ve noticed smoother inference after making that change. Also totally agree on VRAM cleanup between runs, it helps prevent those weird slowdowns.

0 Likes