Dummond D. Slow (mental.delete@this.protozoa.us) on October 9, 2020 9:43 am wrote:
> Doug S (foo.delete@this.bar.bar) on October 9, 2020 8:11 am wrote:
> > Dummond D. Slow (mental.delete@this.protozoa.us) on October 8, 2020 4:39 pm wrote:
> > > Wilco (wilco.dijkstra.delete@this.ntlworld.com) on October 8, 2020 3:02 pm wrote:
> > > > Groo (charlie.delete@this.semiaccurate.com) on October 8, 2020 2:48 pm wrote:
> > > > > itsmydamnation (no.delete@this.way.com) on October 8, 2020 2:38 pm wrote:
> > > > >
> > > > > > While its all ARM level marchitecture for performance increase ( :) /s
> > > > > > dont kill me ) AMD actually sold themselves short in some regards.
> > > > > >
> > > > > > What i mean by that is they game tested @ 1080p not 720p which increases how often they are GPU
> > > > > > bound in a frame. LOL and CS:GO are by far the lightest GPU loads and also low CPU thread count
> > > > > > and they show ~50% performance increase. That extra ~30 points of performance over base average
> > > > > > IPC increase is going to be largely from low thread counts having access to 32mb of L3. If they
> > > > > > game tested @ 720p and picked some more notorious low thread count games ( CIV5, cities skylines,
> > > > > > etc) they could probably skew there gaming performance even higher from 26% perf improvement.
> > > > > >
> > > > > > Worth noting this is the Same 7nm process as Zen2 and thus no EUV until Zen4.
> > > > >
> > > > > They gave out numbers at 4K as well, just not in the video. I won't repost or
> > > > > quote AMD's benchmark numbers until they get their heads out of their ass and
> > > > > properly disclose test setups like Intel does. That said they are out there.
> > > > >
> > > > > Also the process is NOT the same as Zen 2, it is still a non-EUV 7nm but a different variant than
> > > > > before. AMD obviously didn't go into the weeds here when I asked, and TSMC's ever shifting naming
> > > > > conventions make things worse every time they update things, but it is not the same as it was.
> > > >
> > > > N7P is the only non-EUV 7nm process that is not the original N7.
> > > >
> > > > Wilco
> > >
> > > AnandTech says it is N7 with some extra mid-lifecycle tweaks
> > > that are still in the envelope of the original tech.
> > > They rolled those in for the Ryzen 9 3900XT/R7 3800XT/R7 3600XT processors (bot not the original
> > > 3000/3000X SKUs) and supposedly the process for Ryzen 5000 is exactly the same as that.
> > > Of course, AnandTech could make a mistake here, it sometimes happens.
> >
> >
> > So perhaps sort of an N7PP, where the N7P process has been tweaked further.
>
> No, I am pretty sure it is supposed to be a lesser improvement to N7 than what full N7P is/was supposed
> to be. Not a new process, just minor tuning inside the constraints of the original N7 product.
>
> The process shouldn't be nearly as big as improvement over N7 as N7P would be.


Why would it be a lesser improvement? N7P was not a new process, it was simply a tuned N7. What Intel would call a "+" - and what Intel used to deliver without even formally announcing it as they'd tweak the process throughout its life. We just didn't care in the past because there was always a brand new process coming every two years like clockwork so the tweaks were irrelevant unless you were an overclocker.

It isn't like N7P costs TSMC more to deliver, and they will save money using some sort of N7 1/2 P.