Speculation: Ryzen 4000 series/Zen 3

Page 86 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DisEnchantment

Golden Member
Mar 3, 2017
1,590
5,722
136
You are right but they strangely put out this slide where the big dies have even better yield (or less defects) than the small ones or i'm reading it incorrectly.

vlsi-2019-tsmc-n7-yield-2.jpg

I have seen this slide before, but I have never found any value how big 'Large Die' is. Anyone knows?
 

Olikan

Platinum Member
Sep 23, 2011
2,023
275
126
I'm not familiar with parametric yields?
It's the probability of your chip to reach the target clocks/power.... chipmakers counter this with binning

Zen1 to zen1+ is a great example of parametric yields improvement. Since there wasn't needed the 28xx series
 
  • Like
Reactions: moinmoin and soresu

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
You are right but they strangely put out this slide where the big dies have even better yield (or less defects) than the small ones or i'm reading it incorrectly.

vlsi-2019-tsmc-n7-yield-2.jpg
I think we're looking at two variations of the 7nm process here. Mobile, where density and low power is supreme and HPC, (aka N7 Large Die), where clocks have more value. Probably trying to be as dense as possible introduces more defects than slacker density rules. EUV with the fantastic line resolution it brings will change things for a while.

If I'm correct, it really isn't larger die but the process used for HPC, so Ryzen CPU chiplets should be considered as large die. Anyone can confirm?
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,620
136
This is a pipe dream, IMO. I mean, it looks pretty- very orthogonal. Looks like something drawn up by marketing, not engineering (you’d think they’d learn from tick-tock). As feature size decreases, neat little patterns break down.
I agree it looks like a pipe dream right now, we already discussed this leak before. But that's solely Intel's problem to solve. AMD problem is to be prepared for wherever its competition might be at a given time, pipe dream or not.
 
  • Like
Reactions: spursindonesia

jamescox

Senior member
Nov 11, 2009
637
1,103
136
Compared to Matisse, it's more the all core frequency that's the difference.
If intel wins just because of the higher clock, then stop talking about IPC. If intel had significantly better IPC, then they wouldn’t need higher clocks to win. Conversely, if you have something that scales very well with clock speed, then that means it isn’t dependent on memory latency. It has to be very cacheable for that to work. Regardless of the reasons, intel winning at super low quality isn’t actually important. I have seen quite a few benchmarks where they are already flat, even at 1080p.
 

jamescox

Senior member
Nov 11, 2009
637
1,103
136
According to the pic's source, it's 250mm squared or higher.

There's also this article where the graph is very similar (colors changed and doesn't have the "large die" bit) but, most importantly, gives us N7's defect density.

View attachment 17167 View attachment 17166
Don’t know what the second one is, but the first one is defect density reduction over time, not absolute defect density. That sounds like a good way to make it look good when it isn’t. I would expect them to have good and fast reduction in defect density since it probably started out absolutely terrible. Also, if that is 20 nm planar? Did we ever get a big gpu on 20 nm? I thought they skipped that node.

We haven’t really gotten much in the way of large GPUs on 7 nm. Nvidia seems to have released a set of higher clocked rebrands at 12 nm rather than 7 nm parts. The mi60 is 331 square mm, but that is a low volume and very expensive part. Nvidia Ampere should be coming eventually, which will give us some info. I don’t think the yields are that good, so smaller chips will still be preferred going forward.

Edit: actually how available are AMD 7 nm GPUs?
 
Last edited:
  • Like
Reactions: lightmanek

H T C

Senior member
Nov 7, 2018
549
395
136
Don’t know what the second one is, but the first one is defect density reduction over time, not absolute defect density. That sounds like a good way to make it look good when it isn’t. I would expect them to have good and fast reduction in defect density since it probably started out absolutely terrible. Also, if that is 20 nm planar? Did we ever get a big gpu on 20 nm? I thought they skipped that node.

We haven’t really gotten much in the way of large GPUs on 7 nm. Nvidia seems to have released a set of higher clocked rebrands at 12 nm rather than 7 nm parts. The mi60 is 331 square mm, but that is a low volume and very expensive part. Nvidia Ampere should be coming eventually, which will give us some info. I don’t think the yields are that good, so smaller chips will still be preferred going forward.

Edit: actually how available are AMD 7 nm GPUs?

If the 1st pic is, so is the 2nd. Look closer @ the graphs, ignoring the colors (see pic below): they are the same graph, other than that "large die" line and the defect density number.

About the 20nm, that's TSMC's node: do they make 20nm wafers solely for CPUs / GPUs? Don't they have other costumers besides CPU / GPU manufacturers?

tsmc-n7-ramp.png
 

DisEnchantment

Golden Member
Mar 3, 2017
1,590
5,722
136
According to Charlie, some form of X3D will make it to Milan/Zen 3. If there is someone out there that I would trust when it come to these things it would be Charlie.

What is X3D? It is a hybrid of 2.5D packaging, aka interposers/MCMs, and 3D chip stacking, aka HBM/vias. When will it happen? SemiAccurate’s best info says that it will be in Milan, just like we told you last April. Be honest, I’ll bet you thought we were joking back then, right? In any case it sounds like the X3D version will only be in some variants of Milan but will be much more prominent in Genoa.


So it seems we could expect some form of stacking in Zen3.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,620
136
According to Charlie, some form of X3D will make it to Milan/Zen 3. If there is someone out there that I would trust when it come to these things it would be Charlie.


So it seems we could expect some form of stacking in Zen3.
If true AMD will have a track record of revamping their packaging every single Zen gen (discounting the consumer only Zen+ filler).
 
  • Like
Reactions: lightmanek

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
According to Charlie, some form of X3D will make it to Milan/Zen 3. If there is someone out there that I would trust when it come to these things it would be Charlie.




So it seems we could expect some form of stacking in Zen3.

Ehh, I wouldn't trust Charlie too much. He really seems to dislike Intel.
 

etrin

Senior member
Aug 10, 2001
692
5
81
I am confused on release. Last July they were saying Zen3 would be out June 2020. Now there are post stating March 2021.
Is there any news about when that know?
I was going to get one in June but has it shifted a year?
 

inf64

Diamond Member
Mar 11, 2011
3,686
3,958
136
I am confused on release. Last July they were saying Zen3 would be out June 2020. Now there are post stating March 2021.
Is there any news about when that know?
I was going to get one in June but has it shifted a year?

I'm not sure where are you getting that info from? We have just had Analyst Day, they stated Zen3 will be coming in late 2020 (desktop presumably).
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
I'm not sure where are you getting that info from? We have just had Analyst Day, they stated Zen3 will be coming in late 2020 (desktop presumably).
...Zen 3 processors are on track to launch this financial year, giving us a window of anywhere between now and March 2021.
AMD has promised that its Zen 3 architecture be introduced later this year, while consumer CPUs powered by the Zen 3 architecture will arrive in March 2021.
Either way, the fact that we know AMD Zen 3 processors will be hitting the streets by March next year is helpful, even if AMD didn't help narrow the window any further.

¯\_(ツ)_/¯

Seems people are confused about it.
 

Hitman928

Diamond Member
Apr 15, 2012
5,182
7,633
136




¯\_(ツ)_/¯

Seems people are confused about it.

Anandtech got clarification from AMD directly, I have no idea where the consumer products are March 2021 came from, as far as I can tell someone just made that date up?

In this graph, we see that the Zen 3 product here is on the far right, but so is the date – 2021. Does this mean Zen 3 for consumers is coming 2021? We asked AMD to clarify, and were told that we should interpret this as that the range of Zen 3 consumer products, such as desktop CPUs, HEDT CPUs, mobile APUs, and consumer APUs, should all be available by the end of 2021. The company clarified that Zen 3 will hit the consumer market ‘later this year’, meaning late 2020.

 
  • Like
Reactions: lightmanek

amrnuke

Golden Member
Apr 24, 2019
1,181
1,772
136
I am confused on release. Last July they were saying Zen3 would be out June 2020. Now there are post stating March 2021.
Is there any news about when that know?
I was going to get one in June but has it shifted a year?
When did AMD say June 2020?
When did AMD say March 2021?
They are on a 13-15 month cadence on real releases, and a targeted 12-18 month cadence per Mark Papermaster, and Zen2 was released in July 2019. That puts Zen3 in the July 2020 - January 2021 (or more tightly, August 2020 - October 2020) range.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,620
136
Financial year goes until March 2021. But AMD stated Milan (the Ecpyc server chips) is "late this year" whereas Vermeer (the Ryzen desktop chips) is "later this year" which I'd assume is earlier than "late this year". *shrugs*

Edit: See posts below.
 
Last edited:
  • Like
Reactions: lightmanek

Hitman928

Diamond Member
Apr 15, 2012
5,182
7,633
136
Financial year goes until March 2021. But AMD stated Milan (the Ecpyc server chips) is "late this year" whereas Vermeer (the Ryzen desktop chips) is "later this year" which I'd assume is earlier than "late this year". *shrugs*

Where did you get that AMD's fiscal year ends in march?
 

RetroZombie

Senior member
Nov 5, 2019
464
386
96
Maybe:
- Desktop September 2020 (first models)​
- Server November 2020​
- Threadripper January 2021​
- APUs March 2021 (last models)​
?
 

amrnuke

Golden Member
Apr 24, 2019
1,181
1,772
136
Maybe:
- Desktop September 2020 (first models)​
- Server November 2020​
- Threadripper January 2021​
- APUs March 2021 (last models)​
?
That would be reasonable, there may be a case for making Threadripper a pre-Christmas release though it's definitely less holiday-sensitive than desktop.

The most important thing for AMD (IMO) is to get Zen2 Ryzen 4000 mobile CPUs in as many lineups as possible before school starts up again.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
That would be reasonable, there may be a case for making Threadripper a pre-Christmas release though it's definitely less holiday-sensitive than desktop.

The most important thing for AMD (IMO) is to get Zen2 Ryzen 4000 mobile CPUs in as many lineups as possible before school starts up again.
Right. Also, laptop/mobile is the only area that AMD does not currently dominate (except the 1% high end of gamers that want a 9900k for the most fps).

And laptops are very popular these days.