AMD adds support for PyTorch development on select RDNA 3 GPUs with ROCm 5.7

This is fantastic progress. I actually like Nvidia, but anything that levels the playing field and encourages competition here would be a big win.

Related:<p>2023-06-02: <i>Lisa Su saved AMD – Now she wants Nvidia&#x27;s AI crown</i> <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=36164055">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=36164055</a><p>2023-08-09: <i>Making AMD GPUs competitive for LLM inference</i> <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=37066522">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=37066522</a><p>2023-09-26: <i>ROCm is AMD&#x27;s priority, executive says</i> <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=37663194">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=37663194</a><p>2023-10-06: <i>AMD may get across the CUDA moat</i> <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=37793635">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=37793635</a>

&quot;AMD is determined to continue to make AI more accessible to developers and researchers that benefit from a local client-based setup for ML&quot; it seems false. To me AMD is under investing in that regard for many years now.

Meanwhile, on CUDA side, pick any NVidia card, not a couple of selected ones.

Is RocM only working on Linux for now? And if so, it is working alright inside of WSL?

The article mentions 48gb workflows. Presumably two cards attached? What does this look like in the AMD world, I&#x27;m used to what Nvidia offers

ROCm? I hardly KNOWm!<p>I&#x27;m sorry, I know this is a dumb and off-topic comment more appropriate for Reddit, but I couldn&#x27;t resist. If you check my comment history, you&#x27;ll see I don&#x27;t usually do this. Please forgive.