[tric]

> wouldn't it be a duopoly considering that AMD is also a big player?

I don't think GPUs are commoditized. You can't swap a Nvida GPU with a AMD GPU, and get the same performance/results.

[midhir]

AMD seem to be catching up quickly lately. I'm running Stable Diffusion, Llama-2, and Pytorch on a 7900XTX right now. Getting it up and running even on an unsupported Linux distro is relatively straightforward. Details for Arch are here: https://gitlab.com/-/snippets/2584462

The HIP interface even has almost exact interoperability with CUDA, so you don't have to rewrite your code.

[ABeeSea]

Inference and training are not the same things. AMD has basically no market share in training.

[zirgs]

Now try doing the same on Windows.