Speculation: Ryzen 4000 series/Zen 3

Page 113 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
N7P was superior to N6.
N6 can be up to N5 in perf/power in RTO and NTO options. With its compatibility with N7/N7P, it also supports higher volume compared to N7+. More customers are expected to RTO and logic physical redesign NTO, than to go full N7+ redesign.

Samsung's 5LPE 6-track & 7.5-track is on-par with TSMC's N6 6-track & 7.5-track.
 
Last edited:

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
N6 can be up to N5 in perf/power in RTO and NTO options. With its compatibility with N7/N7P, it also supports higher volume compared to N7+. More customers are expected to RTO and logic physical redesign NTO, than to go full N7+ redesign.

Samsung's 5LPE 6-track & 7.5-track is on-par with TSMC's N6 6-track & 7.5-track.

One of these days you will surprise us with a source or drop a link.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
One of these days you will give a source or drop a link.
Fusechip & semiwiki

Most of the stuff added to 5LPE was already in TSMC's N7P. The only thing it lacked was EUV and the enhanced routing efficiency/yield that it provides.

N7+ is a big shrug as most companies who wanted to go on it have dropped it for N6 instead. Technically, more room for AMD, but N7+ and N6 use the same Fab. With N6 to ramp to a higher volume than N7+.

Also, N6 might have + option to support RTOs from N7+ w/ 4 layers to N6(+?) w/ 5 layers and NTOs from 7+ to support even further optimization of the 5-track 7+ option. N6 is suppose to evolve from N5 learning, so it going to N7+ or further than it, is a given.

Also, note there is no discrete N6/6nm node on the website: https://www.tsmc.com/english/dedicatedFoundry/technology/logic.htm
Instead, it is in the N7/7nm section => https://www.tsmc.com/english/dedicatedFoundry/technology/logic.htm#l_7nm_technology
It is completely feasible that "7nm/N7" mentioned can refer to any node in the family; N7, N7P, N7+, N6
 
Last edited:

inf64

Diamond Member
Mar 11, 2011
3,685
3,957
136
Zen3+ is possible for APUs, not a straight GPU-less desktop parts. Zen4 looks to be early 2022 product so Zen3 parts need to hold the forth from Q4 2020 to Q4 2021. I don't see any problem with that, Zen3 will most likely be a superior desktop part to anything intel has in store (in every price bracket). Oh and there is Zen2 to cover the "lower" end of the scale, no matter how absurd that sounds right now.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Zen3+ is possible for APUs, not a straight GPU-less desktop parts. Zen4 looks to be early 2022 product so Zen3 parts need to hold the forth from Q4 2020 to Q4 2021. I don't see any problem with that, Zen3 will most likely be a superior desktop part to anything intel has in store (in every price bracket). Oh and there is Zen2 to cover the "lower" end of the scale, no matter how absurd that sounds right now.
So where did Q4 2020 for Zen3 come from?
 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
Fusechip & semiwiki

Most of the stuff added to 5LPE was already in TSMC's N7P. The only thing it lacked was EUV and the enhanced routing efficiency/yield that it provides.

N7+ is a big shrug as most companies who wanted to go on it have dropped it for N6 instead. Technically, more room for AMD, but N7+ and N6 use the same Fab. With N6 to ramp to a higher volume than N7+.

Also, N6 might have + option to support RTOs from N7+ w/ 4 layers to N6(+?) w/ 5 layers and NTOs from 7+ to support even further optimization of the 5-track 7+ option. N6 is suppose to evolve from N5 learning, so it going to N7+ or further than it, is a given.

Also, note there is no discrete N6/6nm node on the website: https://www.tsmc.com/english/dedicatedFoundry/technology/logic.htm
Instead, it is in the N7/7nm section => https://www.tsmc.com/english/dedicatedFoundry/technology/logic.htm#l_7nm_technology
It is completely feasible that "7nm/N7" mentioned can refer to any node in the family; N7, N7P, N7+, N6

Thanks for the more in depth response.
 
Feb 17, 2020
100
245
116
Nobody wants to use N7+ because it's not compatible with N7 designs.

N6 is compatible with N7's designs and has the same rules, so any company looking to do an iteration of an N7 or N7P product will do it on N6. It's easier to go from N7 to N6 than it is to go from N7 to N7+, and you get more out of it.

N7P is a tuned N7, no major changes.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
Nobody wants to use N7+ because it's not compatible with N7 designs.

N6 is compatible with N7's designs and has the same rules, so any company looking to do an iteration of an N7 or N7P product will do it on N6. It's easier to go from N7 to N6 than it is to go from N7 to N7+, and you get more out of it.
I believe the main reason is actually there will be a N6 option that allows for: N7+ tailored-IP and N7 tailored-IP to co-exist.
N6 with N7 support risk => Q1 2020
N6 with N7+ support after N5 hits a specific volume target => Any time after Q2 2020

With the N7+ support giving an option for IP on N7/N6 to co-exist with IP built for N7+. Basically, N6 is a long-term funnel node fixing the distribution problem of having a 7nm DUV and 7nm EUV node. That are largely incompatible with each other, but with the N6 fix is now compatible with each other.

Early 2019 => N6 only supports N7.
Late 2019 => N6 is a superset of N7+.
TSMC's word => "TSMC's N6 technology provides customers with additional cost-effective benefits while extending the industry-leading power and performance from the 7nm family"
Only N7/N7+ are listed above that, which I believe is what the 7nm family is referring too.

Modes of support of the 6nm node:
{Drive 7nm into 6nm}
-> N7 to N6 retapeout => same draw
-> N7 to N6 new tapeout(New M0 routing) => new draw
{Drive 7nm+ into 6nm}
-> N7+ to N6 retapeout => same draw
-> N7+ to N6 new tapeout(New M0 routing) => new draw
{Support both 7nm&7nm+ and full 6nm designs}
-> Combined or mixed N7+/N7 tapeouts and/or completely new pure N6 tapeout.
 
Last edited:
Feb 17, 2020
100
245
116
I believe the main reason is actually there will be a N6 option that allows for: N7+ tailored-IP and N7 tailored-IP to co-exist.
N6 with N7 support risk => Q1 2020
N6 with N7+ support after N5 hits a specific volume target => Any time after Q2 2020

With the N7+ support giving an option for IP on N7/N6 to co-exist with IP built for N7+. Basically, N6 is a long-term funnel node fixing the distribution problem of having a 7nm DUV and 7nm EUV node. That are largely incompatible with each other, but with the N6 fix is now compatible with each other.

Early 2019 => N6 only supports N7.
Late 2019 => N6 is a superset of N7+.
TSMC's word => "TSMC's N6 technology provides customers with additional cost-effective benefits while extending the industry-leading power and performance from the 7nm family"
Only N7/N7+ are listed above that, which I believe is what the 7nm family is referring too.

Modes of support of the 6nm node:
-> N7 to N6 retapeout => same draw
-> N7 to N6 new tapeout(New M0 routing) => new draw
-> N7+ to N6 retapeout => same draw
-> N7+ to N6 new tapeout(New M0 routing) => new draw
-> Combined or mixed N7+/N7 tapeout and/or completely new pure N6 tapeout.

No, you have it completely wrong. N6 is what N7+ should have been. Going from N7 or N7P to N7+, you need to rework your entire design and flow, which takes time and money. Going from N7 or N7P to N6 is a simple port.

N7+ is a dead end node that's more trouble than it's worth. If companies are going to go through the effort of creating a new design, they're going to N5/N5P, not N7+.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
No, you have it completely wrong. N6 is what N7+ should have been. Going from N7 or N7P to N7+, you need to rework your entire design and flow, which takes time and money. Going from N7 or N7P to N6 is a simple port.

N7+ is a dead end node that's more trouble than it's worth. If companies are going to go through the effort of creating a new design, they're going to N5/N5P, not N7+.
The rework is redesigning for M2=CPP(7nm) to M2=(2/3)CPP(7nm+) and locking it between them. Compared to Samsung's 5LPE which supports both M2=CPP and M2=(2/3)CPP.

For example SiPearl is using Zeus which is a 7nm+ product, but it is being taped out on 6nm.
Ex: "Zeus specifically designed takes advantage of the power and area advantages of the 7nm+ process."
6nm.png

The above is an indicator to me that 7nm+ can also be ported to 6nm.

imho, It is a lot nicer on a roadmap to have 7nm/6nm/5nm compared to 7nm/7nm+/5nm. <== Indirectly referencing the APU/CCD/GPU 7nm/6nm/5nm profile.

Ideal exampe of mixed => 7nm+ Zen3 / 7nm+ RDNA2 IP and 7nm I/O on 6nm for Cezanne. Since, Zen3/RDNA2's node is on a case by case basis.
 
Last edited:

Veradun

Senior member
Jul 29, 2016
564
780
136
No, you have it completely wrong. N6 is what N7+ should have been. Going from N7 or N7P to N7+, you need to rework your entire design and flow, which takes time and money. Going from N7 or N7P to N6 is a simple port.

N7+ is a dead end node that's more trouble than it's worth. If companies are going to go through the effort of creating a new design, they're going to N5/N5P, not N7+.
And this is why TSMC is fading out that naming scheme and calling everything N7 class
 

eek2121

Platinum Member
Aug 2, 2005
2,904
3,906
136
So, yeah, this guy usually just throws a bunch of sh*t up against the wall just to see what sticks.
But, how are you 100% sure that there will be no Zen3+? Got a link, proof, other info?

Cut out with the personal attacks.

I don’t make redacted up, find something I’ve said that has been proven to be “wrong”. The person I replied to claimed there would be Zen 2+.

EDIT: Also, unlike some of you, I don’t live and die by these forums. I have a business to run.

We have a zero tolerance policy for profanity in the tech sub forums.

Iron Woode
Super Moderator
 
Last edited by a moderator:

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,683
136
EDIT: Also, unlike some of you, I don’t live and die by these forums. I have a business to run.
Well, after you're done running the business for today, come back and read again what Ajay wrote. You may be surprised and somewhat embarrassed.

The person I replied to claimed there would be Zen 2+.
The person you replied to was talking about Zen 3+.
 
  • Like
Reactions: spursindonesia

exquisitechar

Senior member
Apr 18, 2017
655
862
136
Man, I wish we got some Zen 3 samples in the wild. Crazy how there hasn’t been a single one spotted yet.
Cut out with the personal attacks.

I don’t make redacted up, find something I’ve said that has been proven to be “wrong”. The person I replied to claimed there would be Zen 2+.

EDIT: Also, unlike some of you, I don’t live and die by these forums. I have a business to run.
He’s talking about Moore’s Law Is Dead, not you, lol.
 
Last edited by a moderator:

amrnuke

Golden Member
Apr 24, 2019
1,181
1,772
136
Cut out with the personal attacks.

I don’t make redacted up, find something I’ve said that has been proven to be “wrong”. The person I replied to claimed there would be Zen 2+.

EDIT: Also, unlike some of you, I don’t live and die by these forums. I have a business to run.
He was agreeing with you...
 
Last edited by a moderator:

amrnuke

Golden Member
Apr 24, 2019
1,181
1,772
136
No, you have it completely wrong. N6 is what N7+ should have been. Going from N7 or N7P to N7+, you need to rework your entire design and flow, which takes time and money. Going from N7 or N7P to N6 is a simple port.

N7+ is a dead end node that's more trouble than it's worth. If companies are going to go through the effort of creating a new design, they're going to N5/N5P, not N7+.
That being said, unless Zen4 is substantially a port of Zen3, then Zen3's node has little bearing on Zen4's node (which is clearly labeled as 5nm).

But since AMD are on N7 now with Zen2, it would be interesting to see what process they choose for Zen3, N6 is possible, I guess, considering it's "N7 class". Though if that's the case, I'm not sure why they labeled it "7nm" instead of "N7". Maybe that would have raised too many eyebrows. Wish we had some leaks!
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
That became a common assumption after AMD's FAD where they stated that Milan is coming in "late 2020" whereas Zen 3 will hit the consumer market "later this year".
Thanks! I read elsewhere that it was later this year. But AT summed it up better
The company clarified that Zen 3 will hit the consumer market ‘later this year’, meaning late 2020.
Now I understand people's comments.
 
  • Like
Reactions: french toast

Valantar

Golden Member
Aug 26, 2014
1,792
508
136
I think they summed it up wrong. :p ‘later this year’ != 'late this year' That can be, but is not necessarily the case.
When a business is talking about a release date in vague terms the best practice is to assume the very latest possible date within the time span given, as businesses talk in CYA language. Which is why "later this year" should be read as "before 2021 (up to and including Dec. 31st 2020)". If it were Q3 they would say Q3, as that would make investors happy and drive up stock prices. Now, December 31st is unlikely due to it missing holiday sales, but late November or early December for Zen 3 hitting stores? Wouldn't surprise me even slightly.
 

maddie

Diamond Member
Jul 18, 2010
4,722
4,627
136
When a business is talking about a release date in vague terms the best practice is to assume the very latest possible date within the time span given, as businesses talk in CYA language. Which is why "later this year" should be read as "before 2021 (up to and including Dec. 31st 2020)". If it were Q3 they would say Q3, as that would make investors happy and drive up stock prices. Now, December 31st is unlikely due to it missing holiday sales, but late November or early December for Zen 3 hitting stores? Wouldn't surprise me even slightly.
Who in this present world situation will state a definite release date many months in advance? So much uncertainty exists.
 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
When a business is talking about a release date in vague terms the best practice is to assume the very latest possible date within the time span given, as businesses talk in CYA language. Which is why "later this year" should be read as "before 2021 (up to and including Dec. 31st 2020)". If it were Q3 they would say Q3, as that would make investors happy and drive up stock prices. Now, December 31st is unlikely due to it missing holiday sales, but late November or early December for Zen 3 hitting stores? Wouldn't surprise me even slightly.

I always get a kick out of it when a company uses H1/H2 instead of Q2/Q4. We know what you're up to.
 

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
When a business is talking about a release date in vague terms the best practice is to assume the very latest possible date within the time span given, as businesses talk in CYA language. Which is why "later this year" should be read as "before 2021 (up to and including Dec. 31st 2020)". If it were Q3 they would say Q3, as that would make investors happy and drive up stock prices. Now, December 31st is unlikely due to it missing holiday sales, but late November or early December for Zen 3 hitting stores? Wouldn't surprise me even slightly.
Sure, but even you should agree that there is a difference between "later this year" (which AMD said regarding consumer Zen 3 chips) and "late this year" (which AMD said for Milan), and summarizing both as "late this year" is reducing the amount of information which may turn out to be wrong.

I assumed they were using a bit of insider info without specifically saying so.
You mean AT has insider info that Milan and Vermeer launch at the same time? Because that's what a conflation of both however fuzzy dates implies.