Adrian (a.delete@this.acm.org) on June 16, 2022 11:43 am wrote:
> --- (---.delete@this.redheron.com) on June 16, 2022 10:08 am wrote:
> > Adrian (a.delete@this.acm.org) on June 16, 2022 9:27 am wrote:
> > > Eric Fink (eric.delete@this.anon.com) on June 16, 2022 5:46 am wrote:
> > > > Adrian (a.delete@this.acm.org) on June 16, 2022 2:09 am wrote:
> > > > > Apple's own presentation of last week showed that the new M2 has only 87% of the single-thread
> > > > > performance of i7-1260P, which is not supposed to be the top of Alder Lake P (that
> > > > > would be i7-1280P, but it does not appear to be available anywhere).
> > > >
> > > > Highest single-core GB5 scores for i7-1280P seems to be around 1800, for i9-12900HK around
> > > > 1890-1900. And I am talking about highest, not average or median scores, so that's the best
> > > > possible case. Would love to see SPEC, but I count find any reviews using the SPEC suite.
> > >
> > >
> > > Any GB5 scores for i7-1280P at around 1800 are just wrong.
> > >
> > > The GB5 database is swamped by bad results obtained on misconfigured Windows computers. Now it has become
> > > much more difficult to identify the correct results than
> > > before, because the database Web site no longer allows
> > > anyone to see the complete benchmark results, for example to check the clock frequencies during the tests,
> > > but it requires a registered account for that. I will not
> > > bother to make an account at them, just to be able
> > > to see the complete benchmark results, in order to be able to filter the fake or wrong results.
> > >
> > > Nevertheless, there is no doubt that 1800 is wrong, because that is the score that
> > > I get on a Zen 3 @ 4.8 GHz (no overclocking, no PBO, slower memory than standard;
> > > Zen 3 with standard memory gets around 1810 @ 4.8 GHz and around 1850 @ 4.9 GHz).
> > >
> > > i7-1280P is Alder Lake @ 4.8 GHz, which must obtain a ST GB5 score higher than Zen 3 by 15% to
> > > 20%, so it should obtain a score around 2100. ST GB5 scales very well with the clock frequency,
> > > so an i9-12900HK @ 5.0 GHz must obtain a score not much less than 2200. For desktop Alder Lake
> > > there are indeed many results in the database around 2200, which must be the right ones.
> >
> > No GB5 processor (averaged in whatever they do it, across the
> > corpus of results) gets close to eve 2100 let alone 2200.
> > https://browser.geekbench.com/processor-benchmarks
> >
> > I'm guessing there aren't enough i7-1260P results to get a solid
> > value on that page, but just scanning the list of result,
> > https://browser.geekbench.com/search?q=i7-1260P,
> > I find it very unlikely that a reliable "average" value for this chip is
> > 2100, given that most numbers I see seem to be clustered around 1670.
> >
>
> Any averages for values in the GB5 database are meaningless.
>
> For any given processor name, there are 4 kinds of results in the database
>
> 1. Correct results, with the computer in the standard configuration and with the benchmark run correctly,
> e.g. not concurrently with other programs whose activity would diminish the scores. In all the cases when
> I have searched the GB5 database, the correct results were no more than 5% of the total number of results.
>
> 2. Bad results, with scores much lower than normal for that CPU, either due to junk software
> running on the test computer, or maybe due to hardware problems, e.g. a completely inadequate
> cooling of the CPU, as frequently encountered in some of the cheapest laptops.
>
> 3. Too high results, obtained on overclocked computers.
>
> 4. Fake results, where the name of the CPU to which the benchmark is attributed
> does not correspond to the real CPU ion which the GB5 benchmark had been run.
>
>
> Any average computed over 1% to 5% of correct results and 95% to 99% of wrong results is meaningless.
>
> The only way to use the GB5 database is to carefully examine every result and filter all the bad results.
>
> One of the main ways to recognize the bad results is to look at the complete test
> results, which are now available only for those with registered accounts, and to
> examine the clock frequencies used by the CPU while running the benchmark.
>
> For single-thread GB5, the recorded clock frequency shall match the maximum turbo frequency
> of that CPU. If it was higher, the result must be excluded due to overclocking,

Note that some processors have actual turbo clocks that are 50-150 MHz above the official boost clock by design (stock, not overclocked), happens with most of the Ryzens. But AMD also allows you to raise boost limits manually as a form of OC, so hard to tell.


> if it was
> lower, it must be excluded due to either bad cooling or junk software interference.
>
>
>
>
>
>
>
>