Speculation: Ryzen 4000 series/Zen 3

[NTMBK]

Holy smokes!

View attachment 21725

If true, BIG.

Now I am wondering if this is the reason for the Ryzen 3K refresh.

Surely any initial run will go into Epyc?

 

[AtenRa]

Surely any initial run will go into Epyc?

Why ?? they make way higher profits from Desktop

 

[itsmydamnation]

no way zen3 is 5nm..... if it is that is the biggest coup of all time. like how the hell did they keep that secret!!!!

 

[Gideon]

I don't doubt that 5nm+ can be ahead of schedule.

But a Ryzen product (of all things AMD could produce) being the first thing designed there ?!? It just doesn't line up with any roadmap they've had.

I mean Super-mega awesome if true, but this thing needs more confirmations before it gets any credibility in my eyes.

AMD already confirmed that Zen 3 is 7nm so unless they pulled their biggest con-job (outright lying to the public and investors at the Investor Day in march) this cannot be for Zen 3.

Anyway, extermely interesting I really do hope that AMD releases something on that node ASAP. I just wouldn't get my hopes up too much.

To me a premium APU for an expensive brand e.g. Apple or Microsoft would make the most sense (as it would be a relatively small chip that could be expensive to produce, those clients wouldn't care) ... but I'll believe it when I see it

EDIT: Spelling (and to note that I might have to eat crow if the following post is true)

 

[uzzi38]

https://twitter.com/x/status/1265956411755790336

But it gets even better.

Zen 3 as a whole retaped from N7+ -> N5P

 

[Gideon]

But it gets even better.

Zen 3 as a whole retaped from N7+ -> N5P

You beat me to it. Mother Of God !

If true, this is an absolutely unprecedented move in the industry AFAIK.

Now I'm really hoping this is true!

 

[uzzi38]

You beat me to it. Mother Of God !

If true, this is an absolutely unprecedented move in the industry AFAIK.

Now I'm really hoping this is true!

I will point out it's best to have some reservations about it. The timelines seem off to me... ramp in Q4 2020 usually means you'll see chips hit the market in Q2 or Q3 2021.

 

[Glo.]

https://twitter.com/x/status/1265956411755790336

But it gets even better.

Zen 3 as a whole retaped from N7+ -> N5P

If they retaped Zen 3 for 5 nm TSMC it puts into toilet any performance speculation about Zen 3. Completely.

 

[itsmydamnation]

maybe this is why the rumored zen3 apu's are so close together? first on 7nm second on 5nm?

 

[Gideon]

I will point out it's best to have some reservations about it. The timelines seem off to me... ramp in Q4 2020 usually means you'll see chips hit the market in Q2 or Q3 2021.

Yeah, to me the Financial Day annonucements are also suspicious. They explicitly said "7nm" on slides in march twice (for both epyc and ryzen), yet for-instance totally omitted the process for CDNA2 or RDNA3.

If there was ever any hope for a Q4 5nm+ ramp, they had to know well before the investor day that they were gonna do it. They could have just ommited the process from all those slides and noone would have suspected anything (yet they would not have lied)

 

[NTMBK]

Interesting... If true, this could free up a bunch more 7nm capacity for AMD. If their Ryzen dies are all coming from the 5+ line, then 7EUV chips can all go into Epyc. Meaning AMD might finally have enough capacity to really compete with Intel's scale.

 

[Gideon]

Also as @Andrei. mentions, this makes little sense for servers. After all Milan has been sampling for a while. since late last year if some rumors are accurate, no-way that was 5nm:

https://twitter.com/x/status/1265964950788415489

Such a change so late would be a very big no-no for server vendors (as this means new lengthy validations, etc and might slip schedules). So this would mean that (at least the initial) Milan Server parts would still need to be improved 7nm.

Yet considering that server would be the platform that would benefit the most from 5nm+ (and one of the easiest to cover expensive production costs) I could see them getting new SKUs later with 5nm chips as well.

 

[exquisitechar]

Have to agree that this makes no sense for many different reasons, but a part of me certainly wishes it to be true somehow.

 

[Arzachel]

Desktop chips *are* server chips for Ryzen and the time frames don't match for a product released late this year. The only way this rumour would make sense for me would be if it's actually Zen3 APUs and the 4000 series bit is a misunderstanding.

Edit: that or Zen+ style update maybe.

 

[LightningZ71]

Is it possible that some of the sampled EPYC parts were 5nm pre-production parts? We know that TSMC has been producing 5nm parts for Apple for some time, and that the 5nm improved nodes have been in "Development" for a while and preproduction for much of the year. If some of the LRIP parts were parts that were being sampled to OEMs over the last year, under the guise of an updated revision, it would meet the timelines.

However, I don't think that's what AMD is doing here. If this is true, it would seem to me that EPYC will be getting improved 7nm parts for Milan, with a "quick" update made available with 5nm parts. As for desktop, we know that AMD has announced desktop Ryzen parts for 4Q2020, if they go volume production with 5nm in 4Q, that means they have been doing LRIP for some time prior to that. If that's the case, and we ALSO believe the rumors that the first Ryzen 3 desktop parts would be "higher end" parts only, it would make sense that those would be expensive 5nm parts to keep supply/demand within reasonable bounds. It also makes sense that a Zen2 refresh would then be able to cover the lower end of the CPU market, and dekstop Renoire would cover the APU and bottom end well enough.

 

[Gideon]

Desktop chips *are* server chips for Ryzen and the time frames don't match for a product released late this year. The only way this rumour would make sense for me would be if it's actually Zen3 APUs and the 4000 series bit is a misunderstanding.

Edit: that or Zen+ style update maybe.

It's particularily confusing when compared to this rumor we had yesterday (APUs being Cezanne @ 7nm in early 2021 and Rembrant @ 6nm in 2022) which would imply it can't really be an APU.

I really want to believe this rumor, It just doesn't add up in many ways (doesn't match the published roadmap, doesn't sit well with what has been released about Milan).

Then again, If AMD did suddenly decide to make this move (also producing 5NP version of the chiplet on top of 7N+ to put extra pressure on Intel) about the only place it would make sense, is indeed desktop Ryzen as it would require minimal validation and could be a good pipe-cleaner. But even then Q4 ramp could mean CES announcement at best (with availability in Q2) not Q4 this year ...

 

[Valantar]

Mass production of silicon starting in Q4. Lead time for a useful number of dice is at least two months from that. Plus binning, packaging, testing, shipping, etc. Shipping from Asia to the US or Europe is normally three weeks. There's also the time necessary to get a working BIOS in place and get it out to motherboard manufacturers, ans then the time they need to validate these chips on their boards. If "Q4" means anything other than October, that makes this an early spring 2021 retail launch at best. I would be absolutely shocked if Ryzen 4000 turned out to be 5nm.

 

[uzzi38]

It's particularily confusing when compared to this rumor we had yesterday (APUs being Cezanne @ 7nm in early 2021 and Rembrant @ 6nm in 2022) which would imply it can't really be an APU.

That one seemed like BS from the onset though, atleast with this one N5P being HVM in Q4 is somewhat believable.

 

[DrMrLordX]

Interesting announcement about TSMC, though what it really means for us remains to be seen. There's too much in play to accurately predict how this development changes AMD's product plans.

 

[Veradun]

I will point out it's best to have some reservations about it. The timelines seem off to me... ramp in Q4 2020 usually means you'll see chips hit the market in Q2 or Q3 2021.

Yup, they either started mass production in Q2 and they mean market availability in Q4, or they're starting mass production in Q4 to have market availability in late Q2 (it doesn't even fit a CES2021 paper launch tbh).

Or it may be bs, or just extrapolation starting fromthe smallest piece of information: "4" and N5P, probably meaning Zen4, not Ryzen 4k.

I must admit all this is pretty funny to watch tho, exciting times :>

 

[NostaSeronx]

I rather trust digitimes over expreview.

 

[Gideon]

A very interesting interview with Lisa Su LIVE:

IR 9:00 AM | Advanced Micro Devices: Dr. Lisa Su, President & CEO (AMD)

May 28, 2020

kvgo.com

Lisa Su just mentioned that she feels very good about what they'll have beyond Renoir. The host explcitly asked about 5nm for future APUs (as AMD skipped those in the Financial Day slides). Lisa dodged the question pretty much, but did mention mentioned that 5nm is a very important node, they will use it heavily, Zen 4 will use it, but it's a bit early to talk about timing.

This was an answer to a quiestion talking specifically about mobile, 5nm and stuff after Renoir ...

One could spin quite a few 5nm conspiracy theories from that

EDITs:

Servers:
26:20 Still on track for 10% share 2020Q2. During second part of 2020 they will transitsion from marketshare target to how much it will be from AMD's revenue (finally!)
29:30 Server is prenominantly CPU (GPUs still minor part)

TSMC:
43:40 Wafer supply continues to be tight, but partnership is great

Interview has ended. Lots of other stuff (and probably things eariler) I missed.

 

[inf64]

They must have started production in Q2 for late Q4 availability. I wonder if Zen3 has >8 cores per CCX now

Crazy AMD, what are you doing! They are going in for a kill it seems.

Edit: Lisa Su @ 4:50 in that linked interview stated that Zen 3 is even better than Zen2. There you go.

 

[ksec]

I could give an argument on both side.

Realistically.

1. 5nm+, as with all other leading edge node, will always be a volume node, and that means it will always be on Low Power first. That is mobile or tablet SoC. Modem etc. High Power used for Desktop or Server generally comes in 6 - 12 months later.

2. We are talking about only 6 months away from mass production. With a 7nm design that is not compatible with 5nm+. If AMD had design it on 5nm, and "move" it to 5nm+ it would have made more sense.

3. Even if TSMC 5nm+ is 2Q ahead of schedule, there is very little chance AMD could have finished its design and testing within the 6 months period.

4. Digitimes has a pathetic track record in any rumours or prediction.

Possibilities

1. TSMC did in fact made 5nm+ ahead of schedule. But no one is willing to pick up those capacity due to design and time constrain. AMD is flexible enough to have those. But it doesn't nesccerily means Desktop will have it. It is simply AMD will be using *something* on 5nm+.

2. 5nm+ is actually designed for some customer intend to launch their product in 2Q 2021. And that customer for some reason has decided / or forced to not use those capacity allotment. ( *Cough* Huawei *Cough* )

3. The amount or orders from 7nm is absolutely insane, despite the delays in console, PS5 and Xbox is some very hyped, cant remember PS4 having anything similar. Amazon and AMD are all doing exceptionally well. Which leads to TSMC recent announcement for further expanding their capacity. Moving more products into 5nm is anther way to improve capacity.

4. This is another leak to those currently on the negotiating table with Samsung. The best Fab in Tech, Schedule, Execution and Volume is still TSMC. You may want to think twice before signing another agreement with Samsung ( *Cough* Nvidia and Qualcomm )

 

[IEC]

Could your memory settings be better than theirs?

Memory has almost no impact on Cinebench scores. Actually, thinking about it, running aggressive memory settings while keeping your CPU at stock TDP would probably hurt your MT scores given that it generally requires more vSOC...

This plus I've seen plenty of reviews where just having a certain motherboard will give better scores.

My money is on he has a better motherboard and better memory than whatever they had configured for that test.

It's a ASUS ROG Strix X470-F. Using DDR4-3200 @ XMP settings.

Cooling also shouldn't matter much given that it's a ST test we're talking about, but it IS under a 360mm AIO for what that's worth.