AMD Zen 2 (Refresh) 3900XT/3800XT/3600XT

PeegeeTips · 29 May 2020 at 14:11

i don't believe they want to push the envelope here, all that is needed is a free process lift that allows them to keep the ASP static on their 'existing' SKU's.

costs them nothing, as the extra performance is supplied by better yield.
maintains revenue, as price doesn't have to drop to compete with Comedylake.
and provides a little performance bump to help ease the delay of the 4000 series.
-----------------
it also frees up valuable wafers for all the next-gen consoles / laptops / low-end desktop, that along with high-end desktop and server would othwerwise saturate their available 7nm supply (but not potential demand!).

which is my way of saying i have a lot of sympathy for the 5nm 4000-series/Milan rumour.

Bossman · 29 May 2020 at 19:04

PeegeeTips said:
Comedylake.

Jamin280672 · 29 May 2020 at 20:34

PeegeeTips said:
i don't believe they want to push the envelope here, all that is needed is a free process lift that allows them to keep the ASP static on their 'existing' SKU's.

costs them nothing, as the extra performance is supplied by better yield.
maintains revenue, as price doesn't have to drop to compete with Comedylake.
and provides a little performance bump to help ease the delay of the 4000 series.
-----------------
it also frees up valuable wafers for all the next-gen consoles / laptops / low-end desktop, that along with high-end desktop and server would othwerwise saturate their available 7nm supply (but not potential demand!).

which is my way of saying i have a lot of sympathy for the 5nm 4000-series/Milan rumour.

I keep saying it, I think 3000XT's are N7P, ive been saying it since 3000XT first suddenly popped up, no one believed me, maybe when AMD finalized them they realized there wasnt a decent enough performance increase to release them as a whole new next gen CPU line (Ryzen 4000), so they've released them as a Zen 2 refresh instead, they had to make a quick decision so close to the release, it would also explain where the XT versions suddenly came from out of the blue and all the R4000 news has stopped for now (except Mobile and APU's), its AMD for god sake, they can do what they want and name them what they want, and then change the roadmap, because 5nm isnt quite ready yet, they have changed it to an announcement in Sept / Oct 2020 with release in April / May 2021 on 5nm.

Sound like the perfect marketing move to me, keeps your audience interested too, and also shows that AMD are not willing to release a whole new line of CPU's, call it new tech for minimal performance increases, like some other CPU manufacturer does, win win for AMD.

Anyone should have been able to figure it out, TSMC always said N7P only brought 7% increase of N7 or 10% less power, thats not enough for AMD to release a whole new line of CPU's to the public and call them next gen, AMD have never been about ripping people off.

So now the slide will look more like this:

LePhuronn · 30 May 2020 at 00:45

12 pages just to say 7nm yields are better now than they were 12 months ago?

andy_mk3 · 30 May 2020 at 00:57

LePhuronn said:
12 pages just to say 7nm yields are better now than they were 12 months ago?

You seem surprised

LePhuronn · 30 May 2020 at 01:07

no idea what said:
You seem surprised

Even by this forum's standards, this one is a bit much

and Christ can I post a lot of waffle

LePhuronn · 30 May 2020 at 01:13

Jamin280672 said:
...no one believed me...

That's probably because you post veritable essays worth of totally incorrect nothingness. Jussayin.

There is nothing special about XT. The node is the same, the arch is the same, it's not even a refresh. It is literally nothing more than the natural progress of yields and silicon quality. You know all those golden sample chiplets that make the 3950X which means it draws less power than the 3900X despite having 4 more cores? Well, those chiplets aren't golden sample any more, they're standard sample.

The XT line is just 125W TDP versions with better chiplets, bumped clocks and a fabric clock boost. It's literally free improvement for AMD with the added bonus of being able to charge a little bit more money and kick Intel in the face once again.

Edit: 105W TDP

Grim5 · 30 May 2020 at 03:31

LePhuronn said:
That's probably because you post veritable essays worth of totally incorrect nothingness. Jussayin.

There is nothing special about XT. The node is the same, the arch is the same, it's not even a refresh. It is literally nothing more than the natural progress of yields and silicon quality. You know all those golden sample chiplets that make the 3950X which means it draws less power than the 3900X despite having 4 more cores? Well, those chiplets aren't golden sample any more, they're standard sample.

The XT line is just 125W TDP versions with better chiplets, bumped clocks and a fabric clock boost. It's literally free improvement for AMD with the added bonus of being able to charge a little bit more money and kick Intel in the face once again.

That's right these XT chips seem to be AMDs equivalent of the 9900KS

Gerard · 30 May 2020 at 03:33

Grim5 said:
That's right these XT chips seem to be AMDs equivalent of the 9900KS

So £1400 for retail, £1500 for the oem?

Jacob Finnis · 30 May 2020 at 06:24

Wonder if this is all they feel they need to do for now while they make 4000 series better..

Troezar · 30 May 2020 at 06:30

Jacob Finnis said:
Wonder if this is all they feel they need to do for now while they make 4000 series better..

I think they know they need to keep pushing hard as Intel will be back and will be good.

LePhuronn · 30 May 2020 at 10:26

Bossman said:
He also likes to draw out his answers. Why take 2 minutes to sum up your review when you can take 10?

Because YouTube don't monetise anything under 10 minutes.

LePhuronn · 30 May 2020 at 10:28

muon said:
Wow. If they can do those base clocks at the same TDP, then they must be highly binned.

No, the XTs are getting a TDP bump. 105W I believe? Besides, is it so hard to believe yields have improved this much? Those golden sample chiplets in the 3950X aren't golden sample any more.

LePhuronn · 30 May 2020 at 10:31

GregI said:
Is there any news on these being the same as the 3300x? i.e. all cores on one chiplet?

That's not how it works.

wookiee87 · 30 May 2020 at 11:07

LePhuronn said:
No, the XTs are getting a TDP bump. 105W I believe? Besides, is it so hard to believe yields have improved this much? Those golden sample chiplets in the 3950X aren't golden sample any more.

some of the 3rd gen lineup were 105w tdp (3800x, 3900x, 3950x) if the new xt's are getting clock bumps then the tdp may be the same as the process is more refined, but AMD could increase to 125w and clock speed could go higher than whats circulating around the web, time will tell when the new cpus land

LePhuronn · 30 May 2020 at 13:55

From what's been leaked already, the XT line are rated at 105W. If that doesn't change some SKUs then fair enough, such is the nature of improved process. I did originally think it was a 125W TDP but I can't recall where I heard that.

Jacky60 · 30 May 2020 at 15:58

KungFuSpaghetti said:
Cool, so we should just about get the clock speeds that the original 3900x touted, instead of the 100+hz below (caveat by saying my 3900x always shows most cores on ccx1 hitting 4.6ghz at some point... just not when I'm actually using them).

Lol, sounds about right from what I've seen but any improvement would be welcome. I hope they understate the boost clock speeds this time and over deliver rather than the opposite as previously. That would improve their image in this department no end.

Jamin280672 · 30 May 2020 at 16:05

LePhuronn said:
Even by this forum's standards, this one is a bit much and Christ can I post a lot of waffle

And some people on this forum are just dumb and like to think they know it all, N7P is refined N7, no design changes, just refined, WOW !!!!! https://www.anandtech.com/show/14687/tsmc-announces-performanceenhanced-7nm-5nm-process-technologies

125w TDP ummm NO!!! TDP isnt changing.

Jacky60 · 30 May 2020 at 16:17

Perfect_Chaos said:
Isn't there a bottleneck in the chips meaning higher clock speed does nothing on the current chips? That's what I've heard/seen from at least 2 different people on youtube, and one of them is a benchmarker[

Cooper said:
I think these boost clocks are pretty meaningless, TBH.

..all that really matters is the sustainable, all core clock speeds whilst under load.

4.7Ghz for a few seconds isn't much good to anyone,really.

Agreed, OR even 4 of 8 or 6 of 12 cores sustained at 4.7Ghz/4.8Ghz indefinitely under load. But it's really the sustained under load bit that needs nailing.

LePhuronn · 30 May 2020 at 16:19

some people are dumb? Fair enough. We all know TSMC have 4 different 7nm-based nodes that are all interchangeable and design compatible with each other. That doesn't mean to say Matisse 2 is moving to a different node. You say other people are dumb, yet you refuse to acknowledge that Matisse 2 could be (and probably just is) nothing more than the benefits of 12 months of node maturity.

And the TDP? Well I originally saw the XT models were bumping to 125W but now I can't find it, plus there's more information coming out which says TDP is 105W, which means 2 SKUs aren't having TDP changes. That aligns with node maturity also.

Everything you're posting is already known and has been discussed, don't start banging on saying people are dumb just because you came to the discussion party late and are retreading old information.