Originally Posted by Homeles
Yes, Nvidia did the computing world a massive favor by pushing CUDA + GPGPU so hard. What I am saying is that CUDA will be eventually phased out in something that is vendor agnostic — OpenCL. With AMD, and now Intel making serious pushes into the GPGPU market, it won't be worth the development effort to code for CUDA, assuming OpenCL provides the access that developers need.
They are already popping up in the HPC world.
CUDA is definitely the more widely adopted API right now, but its use will decline in favor of OpenCL.
Yes, CUDA will eventually be phased out but it will be years. This is especially true on large applications that have been already developed with CUDA. They are so complex and large that it will not be worth the effort to move to a more open platform at this time. With development, it is always a case of cost-benefit of upgrade/redeveloping/rewriting now... or later.
Xeon Phi are currently only avaliable to a few special projects and researchers contacted by Intel. Xeon Phi are not avalaible for purchased..... Believe me, I inquired about evaluating one two months ago.
Originally Posted by GingerJohn
Correct me if I am wrong, but are these cards more often than not bought for a single purpose?
With gaming cards we expect one card (or multiple similar cards) to play all the games we throw at it. OK, so if you are fixated on one particular game you might buy the card that does beter in that game but for the most part we expect them to be flexible.
With the professional cards you will mostly be buying them to run one or maybe two similar programs. As such of course you are going to choose the card that does best at that application. As someone said previously, for CAD pople will buy nVidia cards, for certain media applications AMD cards are better.
Therefore the premise of the article, that AMD can't compete with nVidia, is false: They both compete in different segments of the market.
You wouldn't buy a tractor to drive down the highway, and you wouldn't buy a Porsche to plough a field.
With HPC, developers are writing custom code usually optimized for a specific platform.
They compete in the same market.... it's just that their different architecture and drivers provide better performance/development in certain areas more than others.