David Kanter (dkanter.delete@this.realworldtech.com) on February 4, 2015 10:56 am wrote:
> Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on February 3, 2015 2:39 pm wrote:
> > Maynard Handley (name99.delete@this.name99.org) on February 3, 2015 12:36 pm wrote:
> > > http://www.businesswire.com/news/home/20150203006487/en/ARM-Sets-Standard-Premium-Mobile-Experience#.VNEfWELSm-U
> > >
> > > Looks like it will be slightly higher than Apple A8 performance (though,
> > > of course, details are vague till we see the tech blog right ups).
> >
> > A ~2.0GHz Exynos 7420 already equals A8 on integer and memory, being only behind on FP.
>
> Yet again, Geekbench is basically useless, please stop citing it :)

Which benchmark is more useful then? Don't mention SPEC as that is worse in many ways.

> > So it's obvious A72 at 2.5GHz will beat A8 by a huge margin. I guess it will match Denver
> > and Apple A9 on single threaded int/fp/mem performance - it's going to be a close race.
>
> 1. I don't think beating a power-constrained 28nm/20nm TSMC
> processor using a 16FF+ processor is at all impressive.

Just beating is not impressive, but beating by a huge factor is. These are far bigger gains than Intel got when moving to their FinFet process.

> 2. I don't think we can infer anything about memory performance, since
> that is a function of the SOC fabric and memory controller.

The improved fabric bandwidth and efficiency should increase memory system performance assuming the same memory. However if one uses all 4 memory channels then you could see far larger gains.

> 3. Do you want to put a quantitative estimate behind "huge margin"?

Like I said, I'd expect A72 to match Denver (both on SPEC and GB), so about 50%.

Wilco