OpenCL comparisons are not the right ones to make
It is interesting to note that most pro-Iris-pro readers have kept on emphasising Iris-pro's superior OpenCL performance over Nvidia without realising the history and nuances of general purpose GPU computing. Briefly, Nvidia was the first to introduce it through CUDA. Being the first mover it essentially monopolised the space in the early years, and even today CUDA is a far more mature, advanced and popular implementation compared to OpenCL. That is unfortunate, as CUDA is essentially tied to nVidia, while the very raison d'etre of OpenCL was to be vendor agnostic, and basically give AMD and Intel a chance too. Hence the increase in recent OpenCL momentum to align with Intel's entry into serious GPUs. Nevertheless, despite recent improvements and adoption of OpenCL in industry, which I support, the fact is that CUDA is just better. The reason is simple. CUDA is tied to nVidia architecture and can access specialised GPU-specific technologies, while OpenCL is all about portability, which means it cannot leverage many of the GPU-specific tech.
Engineering applications like Matlab, Solidworks, PRo-E, AutoCAD etc - no question about it - they started implementing CUDA in their code years ago and OpenCL only recently. Same goes for the likes of Photoshop, ZBrush, Maya, Modo, Softimage, Cinema4D. One of the main advantages of CUDA from the developer's perspective is also that it is much easier to code with it compared to OpenCL. This helps ensure its continued popularity.
Is Nvidia really interested in advancing OpenCL? It means allowing Intel and AMD to compete on the same platform. Why would they do it when CUDA is already more mature, more advanced, more popular, better implemented in GPU-accelerated applications, can squeeze much more out of nVidia cards than OpenCL, and locks the software vendors into preferring their architecture?
So if you are going to compare general purpose GPU computing, compare Iris Pro's OpenCL performance with nVidia's CUDA performance. It will then hit you like a brick how big a gap is between the two. OpenCL comparisons of the two are misleading and don't mean anything to me.
I am happy that OpenCL is catching up which increases competition in this space, but it is definitely not there yet. And if nVidia plays its cards right (and they've been messing up quite often so who knows), then CUDA will always be better, and will always remain more popular in professional applications.
And don't get me started on OpenGL...
