[doix]

Yeah, I really feel like AMD is struggling with the software aspect. Even back when they were ATI and AMD bought them, the ATI drivers were garbage compared to Nvidia (from my PC gaming experience). After a few AMD and ATI cards, I just accepted the Nvidia tax, where my cards are more expensive and on paper worse, but in practice worked better.

I'm really surprised AMD isn't throwing a whole bunch of money on emulating CUDA. If they could "just" make CUDA work on AMD cards, it feels like Nvidia's position would be severely weakened.

Kind of like how Valve invested heavily into Proton and now gaming on Linux is pretty much fine.

[luyu_wu]

I'm not sure emulating CUDA would be legal, you can look at ZLUDA as an example. It was originally funded by AMD, but got cut for what I presume would be legal reasons. ZLUDA does work amazingly well though from my experience!

[mook]

AMD also doesn't understand that CUDA got big because they worked on cheap consumer cards; once things were working people got interested in expensive specialized cards. Their stack is still focused on the high end only, but there's no ecosystem to support it.

[rcleveng]

To me, this is the most important point and what AMD is missing out from their current strategy. I can take an off the shelf, easy to get 4070 or 4080 and use it with CUDA to learn.

AMD's strategy for people wanting to learn, is basically no strategy.

It's always been the software holding them back, still is, need to invest in the ecosystem and not just the things easy to justify as a revenue driver.

[pjmlp]

That is what ROMc and HIP were supposed to be somehow, but even that isn't really CUDA, as in the polyglot programming language environment, with C, C++ and Fortran first, plus others, followed by Python JIT, libraries, IDE, and a GPU graphical debugger.