Adrian (a.delete@this.acm.org) on November 24, 2020 10:51 am wrote:
> juanrga (noemail.delete@this.juanrga.com) on November 23, 2020 12:16 pm wrote:
> >
> > Take the 1800X as example. Stock RAM is 2667MHz and in Zen µarch MEMCLK and DFICLK
> > are tied, which implies when you overclock the RAM, you are overclocking IF. This means
> > independently of the core clock the CPU is no more in stock settings. In this review
> > memory and IF were overclocked by 20% increasing IPC over stock configuration.
> >
>
>
> I do not understand what you dislike about this.
>
> The target of those benchmarks was precisely to not run the CPUs at their stock configuration, but in a configuration
> where everything external to the cores was identical, including the memory clock frequencies.

I have just explained above this is not true. In Zen architecture the memory clock is
tied to the CPU clock for the interconnect. So when you overclock the RAM, you are automatically overclocking the CPU, reducing core-core and core-cache latencies and increasing the IPC.

A proper measurement of IPC has to keep everything on stocks settings, if the goal is to compare microarchitectures. If the goal, insteadm is pushing the chips to the limit then all the chips in the review would be overclocked. Comparing overclocked chips to chips not overclocked is both biased and useless.

It has been a common practice in reviews since Zen launch, to compare Zen chips with the interconnect overclocked with Intel models running on stock settings. Most reviews are biased and useless, because the IPC of the Intel chips increases by ~7% or ~15% (depending of the model) when overclocking the interconnect.