Adrian (a.delete@this.acm.org) on November 17, 2020 12:52 pm wrote:
> Adrian (a.delete@this.acm.org) on November 17, 2020 12:41 pm wrote:
> > sr (nobody.delete@this.nowhere.com) on November 17, 2020 11:39 am wrote:
> > > Dummond D. Slow (mental.delete@this.protozoa.us) on November 16, 2020 6:22 pm wrote:
> > >
> > > > Uh? Whether the MT score is good or bad, that absolutely depyends on the power consumed.
> > > > If it does better then say the octocore Ryzen 4800U (probably most efficient laptop chip for this task on
> > > > x86 side, ATM) in the particular power envelope, be it 10W or 15W or whatever, that is a good result.
> > > > If it does worse but at proportionally lower power, also good result.
> > > >
> > > > If the AMD beats M1 at the same or even better power = bad result for Apple.
> > > > And that could all happen with the exact same 7500 points score the M1 achieves. What
> > > > that number means absolutely depends on the power consumed... Not sure what is not
> > > > to understand, performance-per-watt is ultimately the name of the game in these multithreaded
> > > > tasks. It's like with GPUs, ultimately your performance is TDP-limited.
> > >
> > > Not against 4800U but against previous Apple laptops power efficiency is about 10
> > > times better at high loads while simultaneously offering better performance:
> > >
> > > yeah-apples-m1-macbook-pro-is-powerful-but-its-the-battery-life-that-will-blow-you-away/
> > >
> > > Wonder if it's good enough or should Apple have stayed with x86.....
> >
> >
> >
> >
> > Of course the new Apple computers with M1 are much better than what they are
> > replacing (except for the lower memory capacity and number of I/O ports).
> >
> > However, that was an extremely low bar to pass, because Apple computers were notorious
> > in recent years for their low performance compared to alternatives.
> >
> >
> > The comparisons with the older Apple products are useful for the Apple customers, to insure them that an
> > upgrade is desirable, but they are completely useless for those who currently are not Apple customers.
> >
> > For the latter, only the comparisons with Intel Tiger Lake, AMD Renoir and AMD Zen 3 are useful.
> >
> > Compared to Intel Tiger Lake and AMD Zen 3, the differences in single-thread CPU performance
> > are too small to have any practical consequence, so the right conclusion is that all
> > these 3 processors have approximately the same single-thread performance, but the Apple
> > M1 achieves it at a much lower power consumption and clock frequency.
> >
>
>
> I want to add that when using a processor with only 4 cores, the smaller power per core does not
> bring much advantage, because it is a small part of the power consumed by the system, the power consumptions
> for different usage cases measured at Anandtech are very similar with those of a NUC.
>
> For a large CPU with 64 or 128 cores, an M1-like efficiency
> would have a much larger impact on the power consumption.

Not sure if this will be the case in M1's favor. In MT Cinebench, it already loses to Zen 2 (4800U) even when AMD's chip is constrained to a lower power budget of 15W. That's not a good sign, because the 4800U is a process node behind and the core design has already been superseded.

> While for the CPU the large IPC and low clock frequency lead to a large advantage in power consumption
> over alternatives, for the Apple GPU I do not see any trick that they could use to have a very different
> power efficiency than the competition, so at the GPU power I believe that the M1 does not have a
> larger advantage in power efficiency than afforded by the superior 5-nm process.