3 ms·
In the price-sensitive consumer space, price/performance matters a lot. But all the other places, libraries/SDKs/interoperability matters as much or more. Most
by capableweb 4y ago
In the price-sensitive consumer space, price/performance matters a lot. But all the other places, libraries/SDKs/interoperability matters as much or more. Most of all the Stable Diffusion stuff that is appearing is heavily powered by nvidia cores, with AMD support being spotty at best. Same goes for many other areas in AI/ML.
- sangnoir 4y agoAMD compatibility (via ROCM) for Tensorflow and PyTorch are perfectly workable (albeit a little finicky on old AMD cards). Stories of AMD's ML demise are greatly exaggerated.
- capableweb 4y agoSure, for Tensorflow and PyTorch, you can use shims. But there are countless of other examples. I'm not saying it's impossible, but even you have to agree that the tooling is not as mature and extensive as with nvidia?
- sangnoir 4y agoAMD upstream their changes: no need for shims for either library. From my point as a dabbler in the arts - most implementations [1] of ML papers use pytorch, tensorflow or other high-level libraries. I'm yet to see one implemented in CUDA directly. 1. e.g. Huggingface transformers or stable diffusion