Linus B Torvalds (torvalds.delete@this.linux-foundation.org) on April 29, 2017 1:58 pm wrote:
> Mr. Camel (a.delete@this.b.c) on April 28, 2017 1:55 pm wrote:
> > > For example, you can certainly get the same performance with an in-order core if you're willing
> > > to run at twice the frequency and/or a significantly beefier memory subsystem. That's the
> > > "with enough thrust, you can make a pig fly" approach to performance. It's been tried.
> >
> > Who tried that?
>
> Maynard already got in there before I did.
>
> I was basically describing IBM POWER6. It was actually pretty impressive in many ways.
> It was very much the "go for frequency and beefy memory subsystem approach", and it did
> work. And it was very unusual in that a company that had done OoO before (in the same
> architecture) ended up doing in-order in a newer CPU, and higher performance.
>
> But it lived on high frequency. And IBM was back to OoO in POWER7.

Not only on high frequency. POWER6 also had run ahead.