4 ms·
This article doesn’t address the real challenge [in my mind]. Framework support is one thing, but what about the million standalone CUDA kernels that have been
by upbeat_general 3y ago
This article doesn’t address the real challenge [in my mind].
Framework support is one thing, but what about the million standalone CUDA kernels that have been written, especially common in research. Nobody wants to spend time re-writing/porting those, especially when they probably don’t understand the low-level details in the first place.
Not to mention, what is the plan for comprehensive framework support? I’ve experienced the pain of porting models to different hardware architectures where various ops are unsupported. Is it realistic to get full coverage of e.g., PyTorch?
- bdowling 3y agoSomeone could reimplement CUDA for AMD hardware. That would be legal because copying APIs for compatibility purposes is not copyright infringement. (See Google LLC v. Oracle America, Inc., 593 U.S. ___ (2021)). AMD is unlikely to do this, however, because it would commodify their own products under their competitor’s API. A third party could do it though. It may make sense as an open source project.
- blueboo 3y agoResearch kernels mostly turn to ash upon publication anyway. The wheel turns and the next post-doc gives ROCm a try and we move on