Michael S (already5chosen.delete@this.yahoo.com) on August 21, 2018 5:07 am wrote:
> AM (myname4rwt.delete@this.jee-male.com) on August 18, 2018 1:06 pm wrote:
> > Michael S (already5chosen.delete@this.yahoo.com) on August 18, 2018 12:49 pm wrote:
> > >
> > > It's public knowledge that A15 power efficiency was poor and A57 not much better if at all better.
> > > A72 - just a little better than A57.
> > > A9/A14/A73/A75 line had much better efficiency at comparable performance.
> > > A7/A53/A55 had dramatically better efficiency. Not at comparable performance.
> >
> > Public knowledge contains a lot of misinformation talked about and posted around,
> > not everything by far is true. Do you have true or at least trustible data?
>
>
> Turned out that data (you decide about trustworthiness) was already posted in this thread.
>
> https://www.realworldtech.com/forum/?threadid=179436&curpostid=179456
>
> https://images.anandtech.com/doci/12520/speceff1.png
>
> Since you probably don't know much about various ARM SoCs, I will spell it slowly:
>
> SpecInt power efficiency (Lower is better), Absolute score is givent in parenthesis, higher is better
> 9780 (12.70) - Samsung 10LPP, Exynos 9810 PS, ARM Cortex A55*, ARM-Cambridge,UK

No this is M3 downclocked to 1.7GHz so it runs at its most efficient point.

> 12551 (17.69) - Samsung 10LPP, Snapdragon 845, Cryo-280 (tweaked ARM
> Cortex A73), ARM-Sophia Antipolis,France + Qualcomm [Durham,NC,US ?]

You mean tweaked Cortex-A75.

> 13243 (13.00) - TSMC 10FF+, Kirin 970, ARM Cortex A73, ARM-Sophia Antipolis, France
> 13338 (13.59) - Samsung 10LPE, Snapdragon 835, Cryo-280
> 17733 (13.72) - Samsung 10LPE, Exynos 8895, Samsung M2, [Mountain View,CA,US ?]
> 19861 (21.73) - Samsung 10LPP, Exynos 9810, Samsung M3, [Mountain View,CA,US ?]
> 21492 (13.62) - TSMC 16FFC, Kirin 960, ARM Cortex A73, ARM-Sophia Antipolis,France
> 21844 (11.75) - TSMC 16FF, Kirin 955, ARM Cortex A72, ARM-Austin,TX,US
> 22309 (10.87) - Samsung 14LPP, Snapdragon 820, Cryo, Qualcomm [Durham,NC,US ?]
> 25168 (12.41) - Samsung 14LPP, Exynos 8890, Samsung M1, [Mountain View,CA,US ?]
> -------
> * - probably, not 100% sure
>
> At first glance, A72 does not look too badly. But it's misleading, because Kirin 955 and
> Kirin 960 are not measured at the same performance point. Kirin 960 was 16% faster.

So at the same performance level, process and SoC designer Cortex-A73 has about 16% better power efficiency - as you would expect from a new core designed to improve efficiency. It shows that the claim that Cortex-A72 is somehow very inefficient is obviously false.

These numbers are also proof that process scaling continues unabated despite claims to the contrary. Performance went up by >50% while power efficiency almost doubled between 14nm and 10nm. I bet 7nm will show similar gains again.

Wilco