David Kanter (dkanter.delete@this.realworldtech.com) on August 11, 2020 9:58 pm wrote:
> Maynard Handley (name99.delete@this.name99.org) on August 11, 2020 6:09 pm wrote:
> > Dummond D. Slow (mental.delete@this.protozoa.us) on August 11, 2020 5:51 pm wrote:
> > > Maynard Handley (name99.delete@this.name99.org) on August 11, 2020 5:34 pm wrote:
> > > >
> > >
> > > > - Tiger Lake hits around 1150 (not great, but a mobile part). Comparing to an Ice Lake Macbook
> > > > Air (I assume about the target product) it's same ST performance, hopefully better graphics
> > > > or lower power to justify its existence... It does get 4000 rather than 3000 for MT.
> > >
> > >
> > > > So by end of this year we should expect Apple at maybe 1650 (maybe as high as 1800?
> > > > maybe for macs but not phones?), AMD at maybe 1600, Intel probably still at 1400.
> > > >
> > > > That should give you enough to calibrate what's reasonable and what's nonsense.
> > > >
> > >
> > > https://browser.geekbench.com/v5/cpu/3236500 ?
> > >
> >
> > That's definitely a better TL showing :-)
> > Here's how it compares to the one I was referencing:
> >
> > https://browser.geekbench.com/v5/cpu/compare/3236500?baseline=815711
> >
> > Presumably the stepping helps somewhat. Real issue, of course, is what power it takes
> > to hit that 1550 (15W?), and how long it can be sustained in different use cases.
> > Maybe by the end of the year all three are clustered around 1660..1650?
>
> Look at the huge gains in a small number of tests for multi-threaded!
>
> It's very nonuniform.
>

Those are the tests that probably make heaviest use of AVX512, and were definitely the ones scaling most poorly in multi-thread. Indicates that there were probably system power settings and power limits for heavy AVX usages that were not tuned yet in the previous runs.

> David
>