Chester (lamchester.delete@this.gmail.com) on November 23, 2020 2:26 pm wrote:
> Doug S (foo.delete@this.bar.bar) on November 23, 2020 12:03 pm wrote:
> > Chester (lamchester.delete@this.gmail.com) on November 22, 2020 8:33 pm wrote:
> > > There's a reason SPEC isn't popular among review sites. No one cares about it.
> >
> >
> > Wow, so you mean a benchmark you have to pay for, and do some work to get running is less popular than
> > pre-packaged benchmarks you can download for free and are mostly co-branded with commercial products?
>
> I mean, you have to pay for it, and it's flawed (see other post). So no surprise it's not popular.
>
> Sometimes reviewers are willing to pay for a benchmark because it does something useful. Guru3d
> for example, has AIDA64 to run memory synthetics because they don't have an in-house bw/latency
> test, and such microbenchmarks can reveal things about the hardware being tested.
>

It is kind of ironic that you mention them in a discussion on how to properly check hardware and IPC :-)

Guru3D has a historic record of (i) testing Zen hardware in overclocked-mode, comparing it to (ii) engineering samples of Intel hardware on stock settings, and (iii) using one of most biased and outlier workloads to measure IPC. It was typical for Guru3D reviews to publish that Zen IPC was about 5% behind Intel when using all those tricks, whereas any decent review reporting the real IPC was quoting a 10% or even a 15% gap.