Hardware FAQ

First, you can take a look at the information in the ComfyUI repository.

Note

If you are using Windows and want to avoid hassles, currently, there are no alternatives to Nvidia. PyTorch is expected to release a native version for AMD for Windows soon, but until then, Nvidia is the only option.

List of GPUs by usefulness:

  1. Nvidia 4090 24 GB
  2. AMD 7900 XTX 24 GB
  3. Nvidia 3090 24 GB
  4. Nvidia 4080 Super 16 GB
  5. Nvidia 4070 Ti Super 16 GB
  6. AMD RX 7900 XT 20 GB
  7. AMD RX 7900 GRE 16 GB
  8. Nvidia 4060 Ti 16 GB
  9. Nvidia 3060 12 GB

Note

You can also look at any performance tests of hardware for ComfyUI as a reference.

Q: Why are there no AMD cards other than AMD 7900 series on the list?

A: ROCM (Radeon Open Compute) officially supports only these cards.

Q: How much RAM is needed in the system?

A: For normal operation, 32 GB is sufficient, but if you want to handle large resolutions with Supir Scaler Workflow, then 64 GB is recommended.

Q: How to use 2 GPUs?

A: The simplest way is to run 2 workers, each assigned to its own GPU, so they can process tasks in parallel.