ROCm is a software that AMD developed and which is universally acknowledged to be quite buggy and far behind its equivalent by Nvidia called CUDA. My comment had nothing to do with AMD’s hardware or marketshare.
Everyone, including AMD, would be better off if Intel and AMD were working together on an open and cross-vendor standard to counter Nvidia’s CUDA.
The number of devices in use out in the world is a direct correlation to how useful a project like ROCm or CUDA is/could be. More devices means devs are more likely to utilize a specific language or library for a specific use. ROCm is open source and attempting to gain more ground simply by expanding to more devices which are already out there. My response to OP is just illustrating that fact.
Example: Nvidia got an early foothold in the AI/ML game in the datacenter because they were first to platform traction with the CUDA toolkit and inference libraries. It’s horrible to use, but is useful. AMD is now trying to catch up to that by deploying alternative hardware and software that covers most of the same use-case, plus they now have APU and FPGA devices that Nvidia does not. That’s the tldr.
Your comment doesn’t make sense. ROCm is a buggy mess that despite years of working on it AMD hasn’t been able to make work well at all.
Intel’s oneAPI on the other hand is cross-vendor and by all appearances so far is good software that has a real shot at beating CUDA if AMD was not shooting itself in its own leg by riding the dead horse that is ROCm.
I work with the entire CUDA toolkit on a daily basis, and it is also a mess. Nvidia is locked in though, and doesn’t plan any rework anytime soon (you can refer to their own statements on this). Any widespread alternative forces greater competition, and better products as a result.
I’ve never met a single engineer who has worked on any of Intel’s acceleration toolchains, but they are just now getting new devices into the datacenter, so maybe it will gain in popularity.
Not a fanboy post at all. The number of devices that AMD has out in the world is just massive. Why they’d “give up” as OP suggested is beyond me.
ROCm is a software that AMD developed and which is universally acknowledged to be quite buggy and far behind its equivalent by Nvidia called CUDA. My comment had nothing to do with AMD’s hardware or marketshare.
Everyone, including AMD, would be better off if Intel and AMD were working together on an open and cross-vendor standard to counter Nvidia’s CUDA.
See my other response above.
Bruh
What an eloquent argument in response.
we are not arguing, you are just going on and on about amd market share when no one was talking about that. what are you on about?
The number of devices in use out in the world is a direct correlation to how useful a project like ROCm or CUDA is/could be. More devices means devs are more likely to utilize a specific language or library for a specific use. ROCm is open source and attempting to gain more ground simply by expanding to more devices which are already out there. My response to OP is just illustrating that fact.
Example: Nvidia got an early foothold in the AI/ML game in the datacenter because they were first to platform traction with the CUDA toolkit and inference libraries. It’s horrible to use, but is useful. AMD is now trying to catch up to that by deploying alternative hardware and software that covers most of the same use-case, plus they now have APU and FPGA devices that Nvidia does not. That’s the tldr.
Your comment doesn’t make sense. ROCm is a buggy mess that despite years of working on it AMD hasn’t been able to make work well at all.
Intel’s oneAPI on the other hand is cross-vendor and by all appearances so far is good software that has a real shot at beating CUDA if AMD was not shooting itself in its own leg by riding the dead horse that is ROCm.
I work with the entire CUDA toolkit on a daily basis, and it is also a mess. Nvidia is locked in though, and doesn’t plan any rework anytime soon (you can refer to their own statements on this). Any widespread alternative forces greater competition, and better products as a result.
I’ve never met a single engineer who has worked on any of Intel’s acceleration toolchains, but they are just now getting new devices into the datacenter, so maybe it will gain in popularity.
lol, and that’s the argument OP was making; forget about ROCm and jump onboard with OneAPI