Montaray Jack (none.delete@this.none.org) on January 17, 2020 2:58 pm wrote:
> Andrei (andrei.delete@this.anandtech.com) on January 17, 2020 1:47 am wrote:
> > anon³ (nope.delete@this.example.test) on January 17, 2020 12:25 am wrote:
> > > Travis Downs (travis.downs.delete@this.gmail.com) on January 16, 2020 8:20 pm wrote:
> > > > Here's a draft of a blog entry I'll post fully tomorrow:
> > > >
> > > > https://travisdowns.github.io/blog/2020/01/08/avxfreq.html
> > > >
> > > > Feedback, including frank feedback about things that are confusing or pointless, free
> > > > proofreading, whatever is welcome. Reply here or by email or whatever you want.
> > > >
> > > > Please don't repost this anywhere other than RWT for the moment, that's for tomorrow.
> > > >
> > > > Ignore TODOs, I'll fix them tomorrow.
> > >
> > > Impressive! (And I don't say that lightly.)
> > >
> > > There are a couple of typos:
> > >
> > > > We’ve got something to sink our teeth info!
> > >
> > > s/info/into/
> > >
> > > is the first one (I think). I will confess that I got through it once but I don't think I can make it
> > > through twice in one night to dig up where the others were :) It's dense but interesting reading, at least
> > > for the type of person who's been lurking around these parts for ages. Thanks for writing it up.
> > >
> > > Though I will say that the endnotes bother me. I much prefer the style where hovering/tapping the reference
> > > number shows a popup with the text. It's akin to footnotes
> > > vs endnotes in printed material. But changing that
> > > would require messing with web tech, and I'd rather stick to my (hardware) day job, thank-you-very-much.
> > >
> > >
> > > Back on topic: anyone know what happened to AMD's adaptive clocking (as used in Steamroller and described
> > > on the front page here)? I have seen no marketing material
> > > implying that it's in Zen/+/2, but I also haven't
> > > seen any analysis at this level of detail or finer for Zen. It seems like that scheme or a refined version
> > > would be a much more effective solution than Intel's brutish
> > > mess. A single heavy instruction burning 75,000
> > > cycles (at 3GHz, 10us halted plus 10us gated down to 1/4 rate) seems... suboptimal.
> > >
> > > I'm also a bit surprised at how long the Intel FIVR takes to slew (according to the numbers in
> > > the other thread plus your measured results). Sure, it's regulating truly massive currents going
> > > into gigantic capacitances (at least for the time scales involved), but my intuition would have
> > > been that a >100MHz controller could have done a bit better. But probably that's just me being
> > > thrown off by recently designing with CFB amplifiers that slew at over 1000V/us....
> >
> > Adaptive clocking is active and working on Zen.
>
> As long as you aren't using linux on a zen laptop.

What? It's hardware feature, what does Linux have to do with it? Unless it's enabled/disabled at boot for some reason.