AMD GF~7nm ; what got cancelled??

[french toast]

Both AMD and Global foundries have confirmed that AMD was planning a Q4 2018 tapeout of a chip on their 7nm LP process, which has now been cancelled;

"That said, while AMD seems to be handling this well, this isn’t a feint on their part, and that they’ve had to make changes behind the scenes. The company has confirmed to us that they had started development of a chip to be made using GlobalFoundries' 7LP process technology, but this one had not been taped out yet (it was on track for a Q4 tape out, according to GF). So we will be seeing products that weren’t originally intended for TSMC."
https://www.anandtech.com/show/13279/amd-to-fab-7nm-cpus-gpus-at-tsmc

So what has just been cancelled? Zen 2 Rome and Vega 20 we know was on TSMC already, Navi most likely was always on TSMC.
AMD have hinted that there is no change of plan for Ryzen 3xxx, so what chip was being manufactured on Globalfoundries 7nm?...Picasso APU successor for Q4 2019?. PS5 APU?
Interesting, hope 7nm desktop APU has not been cancelled or pushed back into mid 2020.

 

[iBoMbY]

Since CPUs always have been the main business for GloFo, I'm pretty sure they were at least working on a 7nm Zen2. The question is, if there really are plans for two different dies this time, then maybe it was the smaller one mainly for desktop products.

Edit: And the next APU (which has been rumored to launch by the end of this year, or Q1 next year) probably is, and always was, 12nm, and that is still a go. A 7nm APU may be something for end of next year.

 

[NostaSeronx]

Navi was slated for GlobalFoundries.

Navi x0 from Polaris x0 => 232 mm squared (14LPP)
Navi x1 from Polaris x1 => 123 mm squared (14LPP)
Navi x2 from Polaris x2 => 101 mm squared (14LPP)

If you get the gist from 28nm -> 14LPP, on 7LP.
Navi x0 would be >123 mm squared.
Navi x1 would be ~101 mm squared.
Navi x2 would be <90 mm squared.

GlobalFoundries lower capacity would be negligible to Navi.

Vega 20 and it's successor would naturally stay between the 300-700 mm squared region. Which would be more applicable to TSMC's higher capacity. The higher ASP also means more cyclical funding: sell, buy more wafers, sell, buy more wafers, all the while profit.

Other 7LP would be of course the CPU, but that was shifted to TSMC early on. The APU is a maybe, but the tapeout wouldn't have been in Q4 2018.

Q4 2018 tapeout with a then if by Q2 2019 launch indicates Polaris successor. Regardless, Globalfoundries 7LP is enough copy-exact to TSMC's N7, it is silly.

 

[NTMBK]

I'm getting flashbacks to Amur and Nolan getting canned due to 20nm sucking.

 

[NostaSeronx]

I'm getting flashbacks to Amur and Nolan getting canned due to 20nm sucking.

It was because 20nm Bulk at GlobalFoundries was cancelled.

20LPM/14XM/10XM was cancelled for Samsung/GlobalFoundries pact against TSMC on 14LPE/14LPP vs 16FF. This however turned out to be GlobalFoundries backstabbing Samsung in a failed effort to get more customers. It also backfired and lead to only Malta being leading-edge. 20LPM/14XM was going to be at both Dresden and Malta.

GLOBALFOUNDRIES Dresden Module Two LLC & Co. KG
March 2013 – May 2014
-> Participate in the development of components for 20nm LPM-AMD, 20nm LPM-Q, 14nm XM (Extreme Mobility) and 14nm LPX (Low Profile Extension) PDKs.
https://www.eetimes.com/document.asp?doc_id=1261956
-> "First 28-nm then maybe we can move to 20-nm or a larger wafer size."

Technically, Globalfoundries shot themselves in the foot first. This time however could have been prevented if they didn't backstab Samsung. However with the first event, there was a leading edge node down the road. Now there is nothing except TSMC.

 

[DrMrLordX]

Since CPUs always have been the main business for GloFo, I'm pretty sure they were at least working on a 7nm Zen2. The question is, if there really are plans for two different dies this time, then maybe it was the smaller one mainly for desktop products.

Edit: And the next APU (which has been rumored to launch by the end of this year, or Q1 next year) probably is, and always was, 12nm, and that is still a go. A 7nm APU may be something for end of next year.

I would think porting Zen2 over to TSMC would not have been too hard anyway, assuming the CCX design for Rome is close enough to what we'll see in Matisse.

Navi was slated for GlobalFoundries.

Navi was pushed so far back that it's unlikely it was scheduled for a Q4 2018 tapeout anywhere. I remain unconvinced.

 

[NostaSeronx]

Navi was pushed so far back that it's unlikely it was scheduled for a Q4 2018 tapeout anywhere. I remain unconvinced.

Navi Server was pushed back not Navi consumer. Specifically, Navi Sever is Navi10 x2.

Which would generally launch after Navi10 x1. In this case, Navi10-enterpise B-stepping is also a thing.

Consumer Navi for RX -> Q4 2018 tapeout (worst). Stepping A, Ships Q1 2019, Available Q2 2019.
Server Navi for Instinct -> Q2 2019 tapeout (worst). Stepping B, Ships Q3 2019, Available Q4 2019.

Both of the above were slotted for GlobalFoundries only. So, if Amur and Nolan happens again. There will be no Navi whatsoever and Vega 7nm will be re-purposed for consumers.

 

[dark zero]

I'm getting flashbacks to Amur and Nolan getting canned due to 20nm sucking.

Not only that, but nVIDIA's offering too.
20nm was a dissaster and even today is a very bad process...

And.... Seems that AMD has a backup with Samsung. Since GloFo and Samsung process are similar, they might use the latter one in case GloFo fails.

 

[french toast]

I didn't know Navi was on Globalfoundries?
Would have been a great chip.
I don't think it was Matisse, I think the contenders are picasso 7nm successor...(Q3/4 2019 L) or Navi...
I am going with grey hawk.

 

[The Stilt]

If there is a single Zen 2 CPU die design as there was with Zen, then most likely the APU design.
The manufacturing costs of the APUs (especially) need to be kept low as possible, so opting a cheaper manufacturing solution for them would make sense.
Naturally the assumption that GF was the cheaper solution is only speculation, but most likely it was.

 

[PeterScott]

Both AMD and Global foundries have confirmed that AMD was planning a Q4 2018 tapeout of a chip on their 7nm LP process, which has now been cancelled;

You mean, what future design has been moved?

TSMC was already getting both the first 7nm CPU (Epyc Rome) and first 7nm GPU ( Vega Instinct pro).

If Servers and desktop are using different parts this time, then it could be Desktop Ryzen.

If not then the other candidates are smaller consumer oriented GPU, or APU.

Either way, if they haven't even begun tape-out, these are likely LATE 2019 products.

Shifting a design that hasn't even begun tapeout should be no issue at all.

So really a non issue.

If anything it will probably be a speedup. Whatever it is will probably go even faster since they will already have 2 designs under their belt at the target Fab, rather than being a first design a different fab.

 

[french toast]

You mean, what future design has been moved?

TSMC was already getting both the first 7nm CPU (Epyc Rome) and first 7nm GPU ( Vega Instinct pro).

If Servers and desktop are using different parts this time, then it could be Desktop Ryzen.

If not then the other candidates are smaller consumer oriented GPU, or APU.

Either way, if they haven't even begun tape-out, these are likely LATE 2019 products.

Shifting a design that hasn't even begun tapeout should be no issue at all.

So really a non issue.

If anything it will probably be a speedup. Whatever it is will probably go even faster since they will already have 2 designs under their belt at the target Fab, rather than being a first design a different fab.

Yes I alluded to the late 2019 launch probability, which puts it in the APU/ VEGA 20 successor catagory, i doubt it could be a mid 2019 product.

I don't see how it could speed things up by moving fabs? They most probably have begun or finished taping in?...would put them back months and millions of $ to move to TSMC imo.

 

[PeterScott]

Yes I alluded to the late 2019 launch probability, which puts it in the APU/ VEGA 20 successor catagory, i doubt it could be a mid 2019 product.

I don't see how it could speed things up by moving fabs? They most probably have begun or finished taping in?...would put them back months and millions of $ to move to TSMC imo.

It speeds it up because you have zero teething problems with two designs already under your belt.

Especially since it doesn't' sound like they even began tape-out. 7nm has been slipping at GF and no one outside of GF would have greater visibility on any of these issues than AMD.

AMD was not blind sided with this announcement. They have likely been discussing the end of 7nm at GF for weeks, and before that GF was probably delaying them because internally they saw the writing on wall, weeks before that.

When was the last public statement on GF 7nm from either company?

 

[Ajay]

If there is a single Zen 2 CPU die design as there was with Zen, then most likely the APU design.
The manufacturing costs of the APUs (especially) need to be kept low as possible, so opting a cheaper manufacturing solution for them would make sense.
Naturally the assumption that GF was the cheaper solution is only speculation, but most likely it was.

So, it makes sense that the next gen APU would be delayed until:
a) TSMC 7nm volume ramps and wafer prices go down or...
b) Samsung 7nm HVM is online, possibly at a low wafer cost for competitive reasons.

 

[french toast]

It speeds it up because you have zero teething problems with two designs already under your belt.

Especially since it doesn't' sound like they even began tape-out. 7nm has been slipping at GF and no one outside of GF would have greater visibility on any of these issues than AMD.

AMD was not blind sided with this announcement. They have likely been discussing the end of 7nm at GF for weeks, and before that GF was probably delaying them because internally they saw the writing on wall, weeks before that.

When was the last public statement on GF 7nm from either company?

30th of may timeframe;
https://www.anandtech.com/show/12831/globalfoundries-gives-7-nm-capacity-update-mulls-skipping-5-nm

 

[Ajay]

30th of may timeframe;
https://www.anandtech.com/show/12831/globalfoundries-gives-7-nm-capacity-update-mulls-skipping-5-nm

WTH? That's a complete 180 to the original article which was based on a GF press release. Confusing

**Edit: Didn't look at the date, the "update" article is older, duh.

 

[PeterScott]

30th of may timeframe;
https://www.anandtech.com/show/12831/globalfoundries-gives-7-nm-capacity-update-mulls-skipping-5-nm

And it doesn't sound enormously positive. It is talking about skipping the next node, and capacity issues an 7nm, and needing a lot more EUV equipment. The new CEO in March might have been given marching orders to make GF profitable and after a few months on the job he may have assessed that it wasn't going to happen with 7nm, and it might be better to pull the plug.

With AMD's first two 7nm designs already at TSMC and GF signalling it would be later and of limited capacity, it already looked like AMD was moving forward without GF being much of a factor. I really think AMD had an early heads up on this, I really don't see GF getting out of 7nm affecting them negatively.

 

[french toast]

And it doesn't sound enormously positive. It is talking about skipping the next node, and capacity issues an 7nm, and needing a lot more EUV equipment. The new CEO in March might have been given marching orders to make GF profitable and after a few months on the job he may have assessed that it wasn't going to happen with 7nm, and it might be better to pull the plug.

With AMD's first two 7nm designs already at TSMC and GF signalling it would be later and of limited capacity, it already looked like AMD was moving forward without GF being much of a factor. I really think AMD had an early heads up on this, I really don't see GF getting out of 7nm affecting them negatively.

Let's hope not, we all need AMD to provide some competition.

 

[french toast]

WTH? That's a complete 180 to the original article which was based on a GF press release. Confusing

Yea it is quite the shock isn't it.
TSMC 7nm should still be a solid upgrade over '12nm'...

 

[PeterScott]

Let's hope not, we all need AMD to provide some competition.

Also it is possible that they could ship a high end consumer version of this when yield and supply improves, it should be pretty good for gaming too.

 

[french toast]

Also it is possible that they could ship a high end consumer version of this when yield and supply improves, it should be pretty good for gaming too.

I think it could compete with a RTX 2080 and a 1080ti.
But the RTX makes the Vega look like a dinosaur, even if in current gen titles it would be very close.
They would have to charge about $750 for it, I don't think that is possible now, would love it though.

 

[Ajay]

I think it could compete with a RTX 2080 and a 1080ti.
But the RTX makes the Vega look like a dinosaur, even if in current gen titles it would be very close.
They would have to charge about $750 for it, I don't think that is possible now, would love it though.

I wonder if a 7nm consumer Vega is coming at all. That slide points to a server/workstation card. Considering that TSMC's 7nm will be in earlier production, at least for larger die sizes like Vega, it might come out only in professional form. I further wonder if a smaller die Polaris follow-on will be the first AMD 7nm consumer GFX chip.

 

[Qwertilot]

Likely. Server is a better fit for Vega's design anyway & a more lucrative market in terms of margins etc. There'll be 7nm server based competition from NV - and sundry others in the AI space you'd imagine - along relatively soon of course, but maybe there's enough market to go round

 

[PeterScott]

I wonder if a 7nm consumer Vega is coming at all. That slide points to a server/workstation card. Considering that TSMC's 7nm will be in earlier production, at least for larger die sizes like Vega, it might come out only in professional form. I further wonder if a smaller die Polaris follow-on will be the first AMD 7nm consumer GFX chip.

Why not 7nm Vega for consumers, if you are building for other markets anyway?

If you can re-purpose a tapeout for a second market that is generally a good thing. It would give them another option, and something higher performance than Vega 64, to make the GPU cupboard look less bare while the work on more consumer oriented designs.

 

[darkswordsman17]

I'm pretty sure it was Navi. For those saying it couldn't because that'd put Navi too late, well unless they've had Navi taped out at TSMC already (which from what I've seen only Vega 20 and Zen 2 have been), then that was the likely quickest route that Navi was going to production.

Unless it was the PS5's APU. But I'm kinda doubtful it was their consumer PC APU as that seems like its lagging behind quite a bit in priority.

Why not 7nm Vega for consumers, if you are building for other markets anyway?

If you can re-purpose a tapeout for a second market that is generally a good thing. It would give them another option, and something higher performance than Vega 64, to make the GPU cupboard look less bare while the work on more consumer oriented designs.

Because it'd be very expensive and might not be much better for gaming than current Vega. I could maybe see the Prosumer line (Frontier?), but I think Vega 20 is tailored for compute/HPC market (its basically current Vega 64 as far as typical GPU aspects - like shader count/CUs - with some specialized stuff, kinda like tensor cores in Volta/Turing) and I'm not sure those are features that would carryover well to consumer workloads like the ray-tracing stuff. Plus, again, it'd be very expensive (like possibly more expensive than Nvidia's RTX expensive and look how that's going over even with expected performance gains and big new features). It would hurt AMD's gaming credibility even further and Vega would become a total joke for average consumers.