New Gadget 2011

While Intel has been talking about the architecture of Many Integrated Core (MIC) at some time, and this time they have demonstrated 50-core chip for the first time at the supercomputing conference SC11. Knights Corner was the name of 50-core chip code that will work as a co-processor for the CPU Intel Xeon E5. Intel states that the University of Texas will build a computer (a measure of computing speed) 10 petaflop by the merger.

Interestingly, Knights Corner project is the result of Intel's Larrabee GPU killer that can be said to fail in an attempt to Intel's foray into this type of computing in which the graphics processor (GPU) began to be used for general purpose computing. However, GPU and Knights Corner actually have much in common: both are large arrays of processing cores, but there are some differences:

Knights Corner is equipped with a core built on the Intel X86 architecture CPU close to normal. Intel believes that with it, they can provide a toolset for developers the better. While this is true to a certain extent, but certainly X86 or not, this heterogeneous architecture is relatively new.

GPU on the other hand tend to be more efficient than the CPU in terms of FLOPS per Watt, even though we were not sure how the Knights Corner in this regard. We have not seen the numbers that are relevant on this matter, so do not jump to conclusions.

In any case, everyone seems to agree that just adding CPUs will not be as efficient as adding many-core chips into the mix. Now which will achieve petaflop limit 1000 (exaflop) and how efficiently the first time ...

Quote Intel than AMD does not want to lose where we had an article yesterday featuring the AMD Opteron 6200 with 16 cores.