Advanced Hardware

Installation Guide

... (keeping existing content)

---

Non-NVIDIA Hardware (AMD / iGPU)

If you are running on an AMD APU (like Strix Halo) or a dedicated Radeon GPU, you can still achieve acceleration.

Linux / Docker Requirements:

1. Device Passthrough: You must pass the render device to the container.
   • Docker flag: --device /dev/dri:/dev/dri
2. Software Backend: You must use a version of PyTorch compiled for ROCm. The standard Eyeris Docker image is currently optimized for CUDA, but can be manually adapted for ROCm by changing the torch installation in the requirements.txt.

Strix Halo / High-end APUs:

For powerful iGPUs, ensure your host has the latest kernel (6.10+) to properly support the hardware. In many cases, these APUs are so fast that CPU-only mode is still significantly faster than older dedicated GPUs.