juanrga (nospam.delete@this.juanrga.com) on February 4, 2015 5:23 am wrote:
> Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on February 4, 2015 1:58 am wrote:
> > juanrga (nospam.delete@this.juanrga.com) on February 3, 2015 5:15 pm wrote:
> > > Hugo Décharnes (hdecharn.delete@this.outlook.fr) on February 3, 2015 2:20 pm wrote:
> > > > Cortex-A57 delivers better performances than Cortex-A15, though their micro-architectures
> > > > are–nearly–identical. We'll see if ARM Ltd. reiterates with Cortex-A72.
> > >
> > > More than 80% performance gains over A57.
> >
> > ... but it was clearly stated that this is at the same mobile power
> > budget. So the gains are from improved power efficiency.
> >
> > > ~50% IPC gain over A57, rest from ~30% higher clocks.
> >
> > I didn't read anything that indicated such a huge IPC gain. Peak performance should be much better of
> > course, but much of that will be due to higher frequencies enabled by the improved power efficiency.
> >
> > Wilco
> >
> >
>
> I did read the preliminary analysis here
>
> http://techreport.com/news/27767/arm-unveils-cortex-a72-cpu-mali-t880-graphics-and-more

That analysis is wrong, ignoring the fact that the comparison is at identical power, not at max frequency. The large gain in power efficiency allows the A72 to run at a higher frequency than A57 or A15, even on the same process. That means the 80% benefit is split between frequency gains enabled by lower power on the same process, frequency gains due to FinFet and IPC gains. Let's say the average IPC gain is 20% (probably not far off given the range Peter mentioned), so that means the frequency gain is 50% (1.5 * 1.2 = 1.8, something else Techreport got wrong).

Wilco