juanrga (nospam.delete@this.juanrga.com) on August 16, 2014 11:53 am wrote:
> Paul A. Clayton (paaronclayton.delete@this.gmail.com) on August 16, 2014 10:31 am wrote:
> > David Kanter (dkanter.delete@this.realworldtech.com) on August 15, 2014 9:41 am wrote:
> > [snip]
> > > 4. Marketing people talk about Xeon-class performance, but
> > > then typically end up delivering Xeon E3 performance :)
> >
> > Why not Xeon Phi performance on Xeon E7-appropriate workloads? ☺
> >
> > More seriously "Xeon-class" is a very flexible description even ignoring the breadth of the Xeon
> > brand. What constitutes being in the same class with respect to performance (within 5% best-to-best,
> > within 10% median-to-median, within 20% best-to-worst)? What workloads are being considered?
>
> This has been answered before. By Xeon-class performance companies are
> meaning about 80--90% of the performance. I.e. within 20% at worst.
>
> Concrete models of Xeon E5 used in the comparison against ARM, benchmarks used and
> the scores achieved were also mentioned before, and for both X-Gene and ThunderX.
>
>

Not that these comparisons will be really relevant. Moving target By the time they get on market, Intel will have another Xeon lineup out. (Haswell-EP at best, at worst they'll be dealing with Broadwell-?)

And assumes nothing will break ARM's designs, like scaling, serial code or uncore. (everybody wishes to have perfect multithreaded scaling, no such thing exists for real world. Not even for websites)