AMD Quietly Funded A Drop-In CUDA Implementation Built On ROCm: It's Now Open-Source
AMD Quietly Funded A Drop-In CUDA Implementation Built On ROCm: It's Now Open-Source
This is a big deal as it kills one of the major vendor lock-ins nVidia has. Gives people much more choice in the products they can buy to achieve their tasks.
amd real af for that
"AMD decided this year to discontinue funding the effort and not release it as any software product".
So AMD decided that it wasn't worthwhile and so the developer released it on his own. AMDs decisions are just baffling. You still can't even install Pytorch for rocm on Windows.
So AMD decided that it wasn’t worthwhile and so the developer released it on his own. AMDs decisions are just baffling.
AMD may have done this to avoid legal entanglements. It allows the solution to exist without a full endorsement from AMD and also lets the open source community drive it to where it needs to be as far as features and functionality.
Along with what the other comment says, they could think that the project is in a good position to stand on its own now. It may have needed the funding to get to a good place where people will support it, but I'd bet it's there now.
novideo's existence setting the bar so low that im giving kudos to a traded for-profit company
AMD now has the secret sauce. Now they need an architecture with proper dual-issue, and even a backport of the CUs onto RDNA3 structures for the RDNA4 lineup would be good, or 2025 RDNA5 would also be good.
Oddly enough, I've been trying to build rocm on slackware for the past week. It keeps failing out somewhere and throwing a HIP error that I can't figure out. I wonder if this has anything to do with it.
This is the best summary I could come up with:
While there have been efforts by AMD over the years to make it easier to port codebases targeting NVIDIA's CUDA API to run atop HIP/ROCm, it still requires work on the part of developers.
The tooling has improved such as with HIPIFY to help in auto-generating but it isn't any simple, instant, and guaranteed solution -- especially if striving for optimal performance.
In practice for many real-world workloads, it's a solution for end-users to run CUDA-enabled software without any developer intervention.
Here is more information on this "skunkworks" project that is now available as open-source along with some of my own testing and performance benchmarks of this CUDA implementation built for Radeon GPUs.
For reasons unknown to me, AMD decided this year to discontinue funding the effort and not release it as any software product.
Andrzej Janik reached out and provided access to the new ZLUDA implementation for AMD ROCm to allow me to test it out and benchmark it in advance of today's planned public announcement.
The original article contains 617 words, the summary contains 167 words. Saved 73%. I'm a bot and I'm open source!