Originally Posted by TheBlademaster01
No, the reason to use CUDA is because it is hardware specific, you have more control over the GPU pipeline, has a greater community, active support from nVidia devs, better tool set and is far easier to debug. The OpenCL software is often buggy and difficult to solve, which might get the job done but crashes every so now and then and has suboptimal performance. Adobe moved to OpenCL as well but Radeon HD 7000 cards were broken since release and I'm not even sure they work properly today. Those are the kind of issues you get with OpenCL.
Undoubtedly, the counter effect that CUDA is architecture specific is the fact that you restrict the flexibility in hardware that can be used. For professionals however support is much more of an issue than flexibility of hardware. They just buy nVidia Quadros and get the job done without having to deal with buggy software. I'm pretty sure a lot still use Fermi and Tesla gen Quadros.
The proverbial "issues with OpenCL and AMD" its one of those things that some people will fall back on long after it hasn't been true.
Adobe don't think there is a problem with it with OpenCL or AMD, nor Apple, nor the mass of OpenCL partners i could list.
As i said before, OpenCL has been neglected for pretty much all of its existence, you are right to cite this reason for CUDA adoption, i never disputed that, in-fact said as much my self.
So at the risk of repeating myself, OpenCL is now getting the attention it has always needed and lacked, it has improved dramatically, enough for Adobe and Apple to switch to it, i have no doubt its still not perfect and may not be for a little while yet.
The 'fall back' position will soon begin to tire, very much like that favorite among some "AMD drivers are crap, Nvidia are better" has been completely used up and scraped for all but the ignorant and those in denial.Edited by Abundant Cores - 8/10/13 at 10:18am