• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Speculation: Ryzen 4000 series/Zen 3

Page 174 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hitman928

Platinum Member
Apr 15, 2012
2,777
2,126
136
Milan was sampling last year, if I recall correctly. I think you're misinterpreting the first slide.

edit: now that I think of it, where did you even get that first slide? It isn't from the slide deck I linked.
It's in there, you must have just missed it.

If you go by when AMD starts sampling to partners, then Zen4 in 2021 is a possibility as key partners start to get samples well ahead of launch but even then, you're back to Zen3 being an end of 2020 product with earlier sampling and Zen4 being a 2022 product with earlier sampling, just like their roadmaps show, even though I can see where you and NBC got confused. The roadmap isn't supposed to line up at the ends, the dates at the beginning and end are more like bookends.

The purpose of these road maps was to show partners and investors the time frame they are on for Zen iterations from Zen 1 to Zen 4. Basically they are just saying between 2017 and 2022 they will have Zen 1 - Zen 4 released. That's it. AMD also provided context during the presentation that isn't found in the slides to better understand the roadmaps, such as:

Tom's Hardware said:
The company says the first Zen 3 chips will land during the tail end of 2020, but the chips will be "fully in the market" for all segments by the end of 2021.
For the client roadmap which ends with Zen3 and the year 2021. So there's wiggle room in these roadmaps and they shouldn't be read as hard launch time frames but rather general guidance of availability.
 

Ajay

Diamond Member
Jan 8, 2001
7,260
2,491
136
First Zen3 was canned sometime before 2H18. (Team A pushed first to Zen5)
The final Zen3 was then split into two designs: On a 7nm family node which is a fake Zen3(cut short design via olde team) and on a 5nm family node which is the real Zen3(longer design phase via another team). Zen(A) -> Zen2(B) -> Fake Zen3(B) -> Real Zen3(C)-N5 -> Zen4(C)-N5+ -> Zen5(A+B)-N3

That is the general gist that I have been given.

If AMD didn't have multiple concurrent design teams, Starship(SSP packages) would only be packin' 48-cores max.

Don't get me on Team D & E which is going to appear for 6 & 7; this early on.
Wait a minute. In the 2016/2017 time range, AMD had two design teams and one implementation team. You are trying to tell me now, that in four years, they've expanded to 5 design teams. Nutz.
 

A///

Senior member
Feb 24, 2017
771
516
106

NostaSeronx

Diamond Member
Sep 18, 2011
3,188
691
136
Wait a minute. In the 2016/2017 time range, AMD had two design teams and one implementation team. You are trying to tell me now, that in four years, they've expanded to 5 design teams. Nutz.
My understanding is the teams were always present but they were secondary, not in lead architectural roles. Even when they were it was for canned products.

For example:
India -> Amur core/soc lead
Israel -> Nolan core/soc lead
Boston -> Excavator core/soc lead
Sunnyvale -> Zen core/soc lead
etc..

Basically, AMD is finally running the same as 2010-2014. Except, it is only for a single core lineage this time.

Free from GlobalFoundries means they can be node aggressive and design aggressive.
Apple A12 launches -> AMD starts sampling 7nm.
Apple A13 launches -> AMD starts sampling 7nmP.
Apple A14 launches -> etc.

I don't see AMD running slower than the Mobile market. If there is a node and it has more than 10,000 wspm. Then, AMD will be on that node like a ruthless monster attacking a pesky "teen" in a horror movie.
 
Last edited:

A///

Senior member
Feb 24, 2017
771
516
106
N6 doesn't make anysense. Unless the new 7nm they are using is EUV, they are not design compatible. Given Zen 4 will be out in a year or so, why bother? AMD is also using a customized 5nm from TSMC.
EUV is the plan or the presumed plan. The rumored, unconfirmed Warhol may be a short run on 6nm scraps. There are supposedly some decent performance benefits, and if 6nm stays on AM4, then it'll give last choice AM4 buyer some wiggle room without spending money on a new AM5 build.
Yes, I think they were targeting 12-15 months.
12 to 18 months, actually. Which is why I don't understand people thinking it was a strict yearly cadence. We have this from Papermaster.


Dr. Ian Cutress: So far AMD’s rate of new products is on track to produce a new core almost every year. The roadmaps quite proudly showcase Zen 3 as almost ready, Zen 4 in development, and Zen 5 further out. Is this cadence sustainable?


Mark Papermaster: We’re on a 12-18 month cadence, and we believe that is sustainable. It’s what the industry demands from us.
If I had to guess when parts launch. I'd say a week or two after the announcement with reviews going up in that time frame, and come the 28th, I suspect it'll be a same day launch. Enough time for reviewers to do all the hardware in a 2-3 week period. I wouldn't be surprised if reviewers are getting hardware at the end of this morning or early October. The 28th is a Wednesday. Zen 3 launching a few days before or the weekend preceding it makes sense if GPUs will be available same day or next day. There will be a 2-3 week gap between the 8th and the 28th, in which time Zen 3 reviews by sites such as AnandTech may go up. This gives everyone ample time to review both CPUs and GPUs unlike the Zen 2 launch mess which someone had to remind me about. I had it in my head that only Zen 2 flagships were up for sale, not the whole lineup.

They still stick true to their word when it comes to launching before consoles. I just find the timing of a late October launch a little sketchy. Either they know Nvidia have low numbers, they won't compete at all, or they feel very confident. Or some other random theory that'll probably get stolen from this thread and published on Wccftech as a fact. :D
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
16,371
5,282
136
The desktop roadmap just posted above shows only Zen 3 through 2021.
I give up. I just posted roadmaps showing Zen4 before 2022, but nobody wants to listen. I can't even find the roadmap showing Zen3 through 2022 (dunno where Hitman got it), but the one I can find contradicts that. Oh well!

Where in that article does an AMD official state or a statement quote from AMD guarantee a 12 month cadence plan? Even the charts don't show such a claim.
The roadmap shows 4 releases (Zen1->Zen4) in 5 years (2017->2022). Period. I don't know how much more simple it can get than that.
 

coercitiv

Diamond Member
Jan 24, 2014
3,858
3,852
136
I believe you are misinterpreting those slides. According to your interpretation, in March 2020, AMD was promising they'd go back in time and release the first Zen 3 product in 2019.

Edit: In other words, in that presentation AMD was promising Zen 4 would roll out some time in 2022 which they should have no problem achieving.
I give up. I just posted roadmaps showing Zen4 before 2022, but nobody wants to listen. I can't even find the roadmap showing Zen3 through 2022 (dunno where Hitman got it), but the one I can find contradicts that. Oh well!
Looks like AMD does these slides inconsistently, depending on what they mean to convey: something happening faster or something lasting longer.

On one side we have Zen 3 launch until the end of the year showed on the end of the scale:


On the other side we have AM4 socket support until the beggining of the year showed on the end of the scale:


Therefore, any promise that AMD makes based on a graph and not specific & clear language should always be interpreted in the most favorable way towards AMD marketing.


Hence, marketing graphs are not roadmaps.
 

jamescox

Member
Nov 11, 2009
119
175
116
All AMD has provided in the last three years have been marketing graphs and vague open-to-interpretation comments in interviews. Which I guess is better than mentioning the year and quarter that you will launch a product, only to miss your target. Intel . . . *cough*
Could they have changed their priorities a little to get the next generation Ryzen 5000 mobile chip out sooner? Intel doesn’t seem like they are going to have that competitive of part in the server market anytime soon but gaining market share in the server market takes a lot of time. Many companies have things planned far in advance, so they will not switch for quite a while. If AMD wants to grab a lot more market share in the short term, then they need to take mobile market share. It would be great if they can get both the chiplet based Zen 3 parts and the monolithic Zen 3 based APU out at almost the same time. I don’t know if they had the resources to do that in the past. Perhaps they have more resources available since the designs for console chips were completed. That may explain some of the rumors of Zen 3 being a Ryzen 5000 rather than a 4000 part (Skipping 4000 name for desktop parts).
 
  • Love
Reactions: spursindonesia

A///

Senior member
Feb 24, 2017
771
516
106
I don't know. Original Zen looks like 2018 and not 2017 on some of those slides.

Could they have changed their priorities a little to get the next generation Ryzen 5000 mobile chip out sooner? Intel doesn’t seem like they are going to have that competitive of part in the server market anytime soon but gaining market share in the server market takes a lot of time. Many companies have things planned far in advance, so they will not switch for quite a while. If AMD wants to grab a lot more market share in the short term, then they need to take mobile market share. It would be great if they can get both the chiplet based Zen 3 parts and the monolithic Zen 3 based APU out at almost the same time. I don’t know if they had the resources to do that in the past. Perhaps they have more resources available since the designs for console chips were completed. That may explain some of the rumors of Zen 3 being a Ryzen 5000 rather than a 4000 part (Skipping 4000 name for desktop parts).
Possibly. It's something I threw out there a month or two ago on here. It's possible, but it comes down to how many console parts have been shipped. AMD only needs to supply a few parts. Sony, for example, has a robot aided automated assembly facility for all their consoles. I don't know who or what Microsoft uses. Given timelines, Sony alone wants 10M consoles by March 2021, up from 6M earlier in the year. No idea about Microsoft. I imagine shipments to both companies began a long time ago to meet that demand.

Would Zen 3 mobile and APUs launched sooner than later be good? Yes, especially with Intel rumored to bring out an 8/16 Tigerlake. I say rumored despite them making a PR statement they would, but PR is PR, and they haven't really been good at delivering for a few years now.
 

Timorous

Senior member
Oct 27, 2008
430
217
116
Would Zen 3 mobile and APUs launched sooner than later be good? Yes, especially with Intel rumored to bring out an 8/16 Tigerlake. I say rumored despite them making a PR statement they would, but PR is PR, and they haven't really been good at delivering for a few years now.
Cezanne using Vega rather than RDNA2 seems like the sort of thing AMD would do if they wanted to get a quick refresh of Renoir out there in less than 12 months*.

*Edit as in 12 months from Renoir launch which was July.
 
  • Like
Reactions: spursindonesia

A///

Senior member
Feb 24, 2017
771
516
106
Cezanne using Vega rather than RDNA2 seems like the sort of thing AMD would do if they wanted to get a quick refresh of Renoir out there in less than 12 months*.

*Edit as in 12 months from Renoir launch which was July.
Per their roadmaps, official or leaked. They keep improving Vega, but they might have a RDNA2 for mobile ready and go against the leaks/what they said. Renoir launched in March and May. Renoir is the same codename for their mobile processors as their APUs.
 

Hitman928

Platinum Member
Apr 15, 2012
2,777
2,126
136
Looks like AMD does these slides inconsistently, depending on what they mean to convey: something happening faster or something lasting longer.

On one side we have Zen 3 launch until the end of the year showed on the end of the scale:


On the other side we have AM4 socket support until the beggining of the year showed on the end of the scale:


Therefore, any promise that AMD makes based on a graph and not specific & clear language should always be interpreted in the most favorable way towards AMD marketing.


Hence, marketing graphs are not roadmaps.
We talked about it when the Zen3 compatibility controversy was raging. AMD quietly started to shift their language/context on their roadmaps. As you can see in the fine print of the socket roadmap, it says AM4 socket support until 2020. In the newer roadmaps presented in 2020 they started to use "through" 2020 (or 2021/2022) when discussing the slides. Yes it is a very subtle and confusing change and AMD should make the roadmaps more visually clear to avoid the confusion, but they give the proper context when talking about the roadmaps, it's just that context doesn't get conveyed in the world's largest game of telephone that is the internet.

For future products AMD is clearly making things a bit vague because if they say 4Q2020 availability and then it is available Jan 1 2021 people will freak out about it being delayed and AMD's sky is falling and they can't execute, etc. Just like with Zen3, as the products near completion, they start to give more firm dates for launch.
 
Last edited:

Gideon

Senior member
Nov 27, 2007
972
1,497
136
For future products AMD is clearly things a bit vague so if they say 4Q2020 availability and then it is available Jan 1 2021 people don't freak out about it being delayed and AMD's sky is falling and they can't execute, etc. Just like with Zen3, as the products near completion, they start to give more firm dates for launch.
Considering the outrage about Frank Azor's tweet, I don't blame them ... I mean, people really expected there will be a full announcement in a day with no leaks based on a single tweet?!? (compare to the leaking sieve that was Ampere announcement)

If it's a such a random tweet on a private account, it was bound to be an announcement of an announcement. That's at least what I expected (though I did expected RDNA2 to come a bit sooner, at around the date that Zen 3 reveal is).
 

Hitman928

Platinum Member
Apr 15, 2012
2,777
2,126
136
Considering the outrage about Frank Azor's tweet, I don't blame them ... I mean, people really expected there will be a full announcement in a day with no leaks based on a single tweet?!? (compare to the leaking sieve that was Ampere announcement)

If it's a such a random tweet on a private account, it was bound to be an announcement of an announcement. That's at least what I expected (though I did expected RDNA2 to come a bit sooner, at around the date that Zen 3 reveal is).
Yeah, that's all I expected as well. You don't put out a veiled meaning tweet about something happening the next day if it's a major product announcement. It was always going to be a teaser or at best a launch date with no other info. When real info is ready to be announced, it will come officially, not through an ambiguous tweet on a personal twitter account. People get way too emotionally invested in this stuff, lol.
 

leoneazzurro

Senior member
Jul 26, 2016
256
234
116
Considering the outrage about Frank Azor's tweet, I don't blame them ... I mean, people really expected there will be a full announcement in a day with no leaks based on a single tweet?!? (compare to the leaking sieve that was Ampere announcement)

If it's a such a random tweet on a private account, it was bound to be an announcement of an announcement. That's at least what I expected (though I did expected RDNA2 to come a bit sooner, at around the date that Zen 3 reveal is).
To me it was clearly the announcement of the date presentation of the new GPU architecture and lineup, the most people could expect by that tweet could have been a teaser of the architecture/features and MAYBE some actual performance/watt improvement figures. Expecting more than that it was simply unrealistic.
 

Zepp

Member
May 18, 2019
33
18
41
First Zen3 was canned sometime before 2H18. (Team A pushed first to Zen5)
The final Zen3 was then split into two designs: On a 7nm family node which is a fake Zen3(cut short design via olde team) and on a 5nm family node which is the real Zen3(longer design phase via another team). Zen(A) -> Zen2(B) -> Fake Zen3(B) -> Real Zen3(C)-N5 -> Zen4(C)-N5+ -> Zen5(A+B)-N3
is this the explanation of what happened to 18h and why Zen3 is being called 19h?
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,188
691
136
is this the explanation of what happened to 18h and why Zen3 is being called 19h?
Not really...
Family 18h is the original Zen family.

K18.0 => Zen1
K18.1 => Zen2
K18.2 => Zen3
etc.

Indicating bigger vector-widths. Zen1 = 128-bit FMAs, Zen2 = 256-bit FMAs, Zen3 = 512-bit FMAs.

I believe the switch to 17h and 19h occurred before the launch of Zen and Zen2 and Hygon.

The shift to 17h brought improvements to Zen/Zen2. The shift to 19h brought improvements to Zen3/Zen4.

19h is based on another ARM core. Zen is derived from K12(ARMv8) and Zen3 is derived from an unnamed core(ARMv8.3+SVE). This is also where the SMT4 part technically comes from as it is derived from Vulcan(JK(K7-K8 then @ Apple) for K12/BK(BC-JG then @ Broadcom) for the unnamed core). Given that, the core of 19h should at least have a unified scheduler; inline with Intel's Core architectures and VIA's cores. However, the big split is the issue ports of FPU is more like VIA's core design rather than Broadcom Vulcan/Intel Core.

In general => 17h execution core != 19h execution core.
Retire should be different.
Mappers should be different.
Schedulers should be different.
Issue width might be the same, but execution units will be different.
etc.

If the above isn't the case, it isn't a real Zen3 core. Rather, just a Zen2+ core being fitted into a Zen3 launch. Basically, being Bulldozer to Piledriver, which aren't physically the same cores between Orochi-B0/Orochi-C0.
 
Last edited:

eek2121

Senior member
Aug 2, 2005
654
596
136
*Edit as in 12 months from Renoir launch which was July.
Renoir launched in March. Not July.

EDIT: Cezanne will either launch with Zen 3, or it will launch in March. I suspect they will launch it sooner rather than later so it matches the Zen 3 desktop cadence.

Oh and they will be ready to beat tiger lake. We already know they have working full speed silicon based on benchmark leaks. This implies it is close to release.
 

Ajay

Diamond Member
Jan 8, 2001
7,260
2,491
136
My understanding is the teams were always present but they were secondary, not in lead architectural roles. Even when they were it was for canned products.

For example:
India -> Amur core/soc lead
Israel -> Nolan core/soc lead
Boston -> Excavator core/soc lead
Sunnyvale -> Zen core/soc lead
etc..

Basically, AMD is finally running the same as 2010-2014. Except, it is only for a single core lineage this time.
AMD went through some massive 'down sizing' plus those who left for better opportunities. AMD has noted itself that the Zen team was very small for a mainline CPU project. I seriously doubt there is any significant work CPU work being done besides the three mainline teams - well, maybe four new since AMD is putting more focus on mobile (I haven't read anything on how mobile is being handled). There are other teams working on related technologies for HPC, interconnects, chip stacking, etc.; but these are pretty small teams (like Boxborough, MA).
 

jamescox

Member
Nov 11, 2009
119
175
116
I don't know. Original Zen looks like 2018 and not 2017 on some of those slides.



Possibly. It's something I threw out there a month or two ago on here. It's possible, but it comes down to how many console parts have been shipped. AMD only needs to supply a few parts. Sony, for example, has a robot aided automated assembly facility for all their consoles. I don't know who or what Microsoft uses. Given timelines, Sony alone wants 10M consoles by March 2021, up from 6M earlier in the year. No idea about Microsoft. I imagine shipments to both companies began a long time ago to meet that demand.

Would Zen 3 mobile and APUs launched sooner than later be good? Yes, especially with Intel rumored to bring out an 8/16 Tigerlake. I say rumored despite them making a PR statement they would, but PR is PR, and they haven't really been good at delivering for a few years now.
It seems like a lot of the console chips would have been made already, so there is some possibility that they have some more fab capacity available. The design team that was working on the console parts should have been finished a long time ago, so perhaps they do have resources to get it finished sooner rather than later.
 
  • Like
Reactions: Tlh97 and A///

jamescox

Member
Nov 11, 2009
119
175
116
AMD went through some massive 'down sizing' plus those who left for better opportunities. AMD has noted itself that the Zen team was very small for a mainline CPU project. I seriously doubt there is any significant work CPU work being done besides the three mainline teams - well, maybe four new since AMD is putting more focus on mobile (I haven't read anything on how mobile is being handled). There are other teams working on related technologies for HPC, interconnects, chip stacking, etc.; but these are pretty small teams (like Boxborough, MA).
The mobile parts are almost the same cores; half the cache and maybe some power tweaks, maybe less aggressive prefetch. They just need a new layout to integrate them with the IO bits. They should already have most of the work done. They would have already ported the IO to 7 nm For Zen 2 based APUs.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,188
691
136
http://www.dolphin-ic.com/products/standard-cell/tsmc_5ff_cell.html // Aug 31, 2019
  • 6-track, Ultra High Density (51nm and 57nm poly pitch)
  • 8-track, High Performance & High Density (51nm and 57nm poly pitch)
  • All tracks are available with 6nm channel length
http://www.dolphin-ic.com/products/standard-cell/tsmc_7ff_cell.html // Oct 18, 2018
  • 6-track, Ultra High Density (57nm poly pitch)
  • 7.5-track, High Performance & High Density (57nm or 64nm poly pitch)
  • 9-track, Very High Performance & High Density (57nm poly pitch)
  • Channel Lengths include 8nm and 11nm
It's very interesting that 6-track 5nm has an option with Std. Cells with 57Cx the same as 7nm/7nm+/6nm 57Cx.

Zen2 => Cx=57-nm / M1=57-nm and M2=40nm/44nm
Zen3;
7nm+/5nm 57poly => Cx=57-nm / M1=38-nm // Major difference is M0/M2/M3/Fin Pitch/Fin Length between 7nm/5nm.
5nm 51poly => Cx=51-nm / M1=34-nm

It appears to reflect Synopsys libs on 6nm. So, it should be official in regards to what AMD has access to.
Only showing 6-track.

Pretty sus for AMD to choose October. Which is also the month date for Apple's launch.

Be the first time AMD missed sampling on the same node w/ Apple on the GPU side. Since, returning to TSMC.
Vega20 N7 = 2018 / A12 N7 (MTS-ES Crash McCrashy also sampled in 2018 btw)
Navi1x N7P = 2019 / A13 N7P
Navi2x ? = 2020 / A14 N5
 
Last edited:

A///

Senior member
Feb 24, 2017
771
516
106
It seems like a lot of the console chips would have been made already, so there is some possibility that they have some more fab capacity available. The design team that was working on the console parts should have been finished a long time ago, so perhaps they do have resources to get it finished sooner rather than later.
Exactly. That's what I've been saying. There's a new damning report out there right now that Sony has cut production for the current agreed upon term ending in March 2021 by 4M units. 11M is now the projected rate. Which meant they had at some point increased their projected order from 10M to 17M. This is the first I'm hearing about it because they increased from 6M to 10M a few months ago.

The issue they're stating is yield issues. However, a deeper dive into the article says they're having trouble reaching the clock speeds on the GPU. This isn't surprising. I doubt Microsoft is having such issues as they took a more mainstream and conservative approach to clock speeds for the GPU vs. Sony's 2.2+ Ghz.
 

ASK THE COMMUNITY