PyTorch backend fails on integrated AMD Radeon 780M

[Ayayayaaaayyaaaay]

PyTorch pipeline fails to load on integrated AMD Radeon 780M with RuntimeError exception.

Debug Info

• Windows Version: Windows 11
• GPU Device: AMD Radeon 780M (integrated, Ryzen 7 H255)
• Amuse Version: v3.5.1

Error log:
[WARNING] failed to run offload-arch: binary not found.

Additional context
Custom safetensors models (SDXL) fail to load via PyTorch backend. OnnxRuntime models work fine.

[saddam213]

Unfortunately the only AMD cards that are supported for Amuse PyTorch are the 7000 series GPU's (7900XTX & W7900)

All other AMD cards are limited to OnnxRuntime

For AMD I recommend ComfyUI or AMD Lemonade Server, these will have way better support for AMD.

ComfyUI for AMD:
https://www.amd.com/en/blogs/2026/amd-comfyui-advancing-professional-quality-generative-ai-ryzen-radeon.html

Lemonade Server for AMD
https://lemonade-server.ai/

[massadm]

Unfortunately the only AMD cards that are supported for Amuse PyTorch are the 7000 series GPU's (7900XTX & W7900)

I think many in the community would be interested to know what exactly causes this limitation. Is it tied to a specific architectural component or requirement in the backend? It would be great to understand what makes the 7000 series GPUs uniquely suited for Amuse PyTorch compared to other generations.

[saddam213]

I just cant get other to work, skill issue

[massadm]

PyTorch pipeline fails to load on integrated AMD Radeon 780M with RuntimeError exception

This is a duplicate of issue #31 . Both issues stem from the exact same root cause: trying to run the PyTorch pipeline on unsupported AMD hardware.

[massadm]

I just cant get other to work, skill issue

Let me know if I can help with debugging on my gfx1151.

[saddam213]

I just cant get other to work, skill issue

Let me know if I can help with debugging on my gfx1151.

Careful, Amuse melted my 395+, and my Kraken turns off mid generation, would not recommend

[massadm]

Careful, Amuse melted my 395+, and my Kraken turns off mid generation, would not recommend

Thanks for the warning. Just disable CCDs Turbo Boost in your BIOS if you are experiencing overheating during inference - it works wonders for stabilizing thermals under heavy AI workloads!