http://www.bbc.co.uk/news/technology-15758057 So what's the arguement between CPU and GPU based supercomputers? I understand GPU requires more programming to do a better job but Intel's CPU approach is plug and play. Is that right? (simplified)
It depends on your definition of harder. The main difference is the number of concurrent threads you can have running - on a quad-core CPU, you can run 4 threads, one for each core. Add in hyper-threading, and we get bumped up to 8 (that's best case though). A GPU has many, many more cores - for example the GTX570 has 480 of them. I'm not sure if that correlates 1:1 with the number of possible concurrent threads but it can certainly support a hell of a lot more than a CPU. So really, it depends on the scope of your problem. Do you have a long string of dependent calculations (that is, a string of calculations where the result of the current calculation depends on the result of the former) to compute? Then a CPU, with its higher clock speed will help. Do you have lots and lots of independent calculations? The speed gains from a GPU will be massive. It all depends on how parallel the computation can be made to be.
It's a shall we say ba$tard son of Larrabee, and is only part based on. They are calling this newer tech MIC.
