wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-01-11 14:50:49 +08:00
Code Issues Actions 11 Packages Projects Releases Wiki Activity
55b187768a
ComfyUI/docs/compatibility.md
doctorpangloss 55b187768a update docs
2025-12-26 16:00:18 -08:00

3.8 KiB
Raw Blame History

Hardware & Software Compatibility

This project is rigorously tested on specific hardware and software configurations to ensure stability and performance.

Compatibility Matrix

Linux

Hardware Python CUDA / ROCm PyTorch Torch-TensorRT Container Image Status
NVIDIA RTX 3090 (24GB) 3.12 12.9.1 Latest 2.8.0a0 nvcr.io/nvidia/pytorch:25.06-py3 ✅ Automated
NVIDIA RTX 3090 (24GB) 3.12 12.8.1 Latest 2.7.0a0 nvcr.io/nvidia/pytorch:25.03-py3 (LTS) ✅ Automated
NVIDIA RTX 3090 (24GB) 3.10 12.6.2 Latest 2.5.0a0 nvcr.io/nvidia/pytorch:24.10-py3 ✅ Automated
NVIDIA RTX 3090 (24GB) 3.10 12.3.2 Latest 2.2.0a0 nvcr.io/nvidia/pytorch:23.12-py3 ✅ Automated
AMD RX 7600 (8GB) 3.12 ROCm 7.0 2.7.1 (Nightly) N/A rocm/pytorch:rocm7.0_ubuntu24.04_py3.12_pytorch_release_2.7.1 ✅ Automated

AMD Note: Automated testing for AMD uses a specific nightly build of PyTorch 2.7.1 optimized for RDNA 3 (gfx110X) from https://rocm.nightlies.amd.com/v2/gfx110X-dgpu/.

macOS

Hardware Python Acceleration Status
Apple Silicon (M1/M2/M3) 3.12 MPS (Metal Performance Shaders) ✅ Automated (macOS 14 Runner)

Windows

Windows support is manually verified.

Hardware Python CUDA Drivers PyTorch Status
NVIDIA RTX 3090 3.10 - 3.12 12.8 560+ 2.7, 2.8, 2.9 ✅ Manually Verified

AMD ROCm Support for Other Architectures

You can install ComfyUI with acceleration on other AMD architectures by pointing uv to the correct package index.

Architecture Table

Find your GPU in the table below to determine the correct index URL.

Series Models (Examples) Architecture Index URL
RX 9000 RX 9070 / XT, RX 9060 / XT RDNA 4 (gfx1200, gfx1201) https://rocm.nightlies.amd.com/v2/gfx120X-all/
RX 7000 RX 7900 XTX, 7800 XT, 7600 RDNA 3 (gfx1100, gfx1101, gfx1102) https://rocm.nightlies.amd.com/v2/gfx110X-all/
RX 7000 (M) Radeon 780M (Laptop), 7700S RDNA 3 (gfx1103) https://rocm.nightlies.amd.com/v2/gfx110X-all/
Strix Halo Strix Halo iGPU RDNA 3.5 (gfx1151) https://rocm.nightlies.amd.com/v2/gfx1151/
Instinct MI300A, MI300X CDNA 3 (gfx942) https://rocm.nightlies.amd.com/v2/gfx94X-dcgpu/
Instinct MI350X, MI355X CDNA 4 (gfx950) https://rocm.nightlies.amd.com/v2/gfx950-dcgpu/

Installation Examples

Use uv pip install with the --index-url corresponding to your hardware.

RX 9000 Series (RDNA 4)

uv pip install --index-url https://rocm.nightlies.amd.com/v2/gfx120X-all/ --pre torch torchaudio torchvision

RX 7000 Series (RDNA 3)

uv pip install --index-url https://rocm.nightlies.amd.com/v2/gfx110X-all/ --pre torch torchaudio torchvision

RX 6000 Series (RDNA 2)

uv pip install --torch-backend=auto --pre torch torchvision torchaudio

RX 5000 Series (RDNA 1)

uv pip install --torch-backend=auto --pre torch torchvision torchaudio

Instinct MI300

uv pip install --index-url https://rocm.nightlies.amd.com/v2/gfx94X-dcgpu/ --pre torch torchaudio torchvision

Notes

  • NVIDIA: Automated testing uses official NVIDIA PyTorch containers to ensure compatibility with the latest deep learning stack.
  • AMD: Automated testing targets ROCm 7.0 on RDNA 3 architecture (RX 7000 series).
  • macOS: Tested on macOS 14 runners with Python 3.12 using the mps backend for acceleration.
  • Windows: While not part of the automated CI loop, Windows builds are manually verified against recent PyTorch and CUDA versions on standard consumer hardware.