Eric (eric.kjellen.delete@this.gmail.com) on November 26, 2013 4:05 am wrote:
> Linus Torvalds (torvalds.delete@this.linux-foundation.org) on November 25, 2013 12:29 pm wrote:
> > And the old Atom really was pretty bad at some general-purpose stuff. That VR-zone link
> > says KNL is 72 modified Silvermont cores, so it should be much better in that regard.
> >
> > Linus
>
> What I am mostly interested in is exactly how they are modified with regard to vector processing. The Knights
> Corner cores and those of its predecessors were basically P54C derivatives with an added 512-bit vector
> unit (for 16 DP FMA flops per core and cycle) but as we speculated in a different branch of the thread (what
> is now "3-TFLOPS-DGEMM") 72 Silvermont cores with a single 512-bit vector unit will not be enough to give
> the cited 3 DP peak TFLOPS at a reasonable clock frequency (they would have to run at 2.6 GHz). The question
> that puzzles me is how Intel, if Heise.de's claim is correct that KNL will be based on 72 cores and Justin
> Rattner's words in 2011 that future MIC will be derived from Atom cores means that the KNL cores are Atom
> based, have managed to make the cores execute 32 DP FLOP/cycle (something that would make the calculations
> work out perfectly - it would give 3 DP TFLOPS at a very realistic 1.3 GHz).

One thought that just struck is something analogous to the double-pumped ALUs of the Pentium 4. Could only the vector unit of the KNL core be run at 2.6 GHz and the rest at 1.3 GHz? I'm not sure if this would actually ameliorate the power draw penalty of running a highly parallel processor at a high clock frequency (as the vector units are probably one of the most power hungry structures) and the same bandwidth constraints would apply in any case.