David Kanter (dkanter.delete@this.realworldtech.com) on February 8, 2015 1:39 am wrote:
> Krait gets all this praise for being so power efficient
> but I have yet to see a real comparison with normalized
> > performance on the same process and all that. All I've seen
> > is comparisons where A15 uses more power (sometimes
> > a lot more) while also performing better (sometimes a lot better).
>
> Not on any vector workloads. The A15 had some terrible behavior when doing SIMD.
>
> I repeat it constantly, but perf/W is
> > not linear and these comparisons are meaningless. Reduce the clock speed on the A15 until it matches the
> > performance on the Krait core and then do a fair measurement - of course, this will vary tremendously from
> > benchmark to benchmark, especially since Krait appears to have some big glass jaws.
>
> What do you think the glass jaws of Krait are?
>
> > What we do know is this: despite your remarks that power management is "not so strong" in A57 Qualcomm
> > still preferred to use it instead of a version of Krait that only added ARMv8 support, something
> > I'm sure they could have done in time if it would have given them such an advantage.
>
> I don't think so. I believe Qualcomm is only using ARM
> cores for their APs due to time to market constraints.
>
> David

Agree it was for time to market constraints. From the time they learnt about the Apple A7 would have been too short, they obviously thought they had another six months at least in hand before they needed such a chip. Even without that using an ARM design for the first version is a reasonable choice to make certain all the design bugs are ironed out doing a new architecture.

Personally I expected ARM to do a 64 bit version of the A17 before they did an A57 follow on like he A72, and that any A57 follow on would be aimed at servers and have the new atomic instructions in. So I was wrong on both counts about that.