[Techpowerup] AMD "Zen" CPU Prototypes Tested, "Meet all Expectations"

[DrMrLordX]

Sounds interesting, but I'm gonna wait until it shows up on roadmaps before I pass any judgment. 2015-2016 is looking like 2016 since 2015 doesn't have very many days left.

 

[Dresdenboy]

There are patents mentioning a low power core next to the high performance cores. IIRC this was about x86.

 

[ElFenix]

There are patents mentioning a low power core next to the high performance cores. IIRC this was about x86.

Was wondering if AMD would try some sort of big little scheme. Though might this also be for putting an arm core in the product for trustzone?

 

[TechGod123]

I'm getting more and more excited now

 

[Dresdenboy]

Was wondering if AMD would try some sort of big little scheme. Though might this also be for putting an arm core in the product for trustzone?

The ARM core cannot be utilized the same way. With the news of FinFet Cat cores there are other candidates. They should be 1-2 mm² (w/o L2), while Zen might have ~ 3-4 mm² (w/o L2).

 

[NostaSeronx]

Piledriver module/core tweaked for 22FDX.

32nm PDSOI; (guessing)
w/ LVT -> 34.5 Lg -> Min Lg
w/ RVT -> 42.5 Lg -> Long Lg
w/ HVT -> 50.5 Lg -> Very Long Lg
~18 mm² 2CMT / ~14 mm² 1CMP

22nm FDSOI; (guessing)
w/ sLVT -> 20 Lg + FBB
w/ RVT -> 20-24 Lg + RBB
w/ RVT -> 28 Lg + RBB
<8 ~ <9 mm² 2CMT / <6 ~ <7 mm² 1CMP

Bulldozer's 1CMP -> 2CMT equals ~12% increase of the module size.

 

[TurtleCrusher]

Piledriver module/core tweaked for 22FDX.

32nm PDSOI; (guessing)
w/ LVT -> 34.5 Lg -> Min Lg
w/ RVT -> 42.5 Lg -> Long Lg
w/ HVT -> 50.5 Lg -> Very Long Lg
~18 mm² 2CMT / ~14 mm² 1CMP

22nm FDSOI; (guessing)
w/ sLVT -> 20 Lg + FBB
w/ LVT/RVT -> 20-24 Lg + RBB
w/ RVT -> 28 Lg + RBB
<8 ~ <9 mm² 2CMT / <6 ~ <7 mm² 1CMP

Bulldozer's 1CMP -> 2CMT equals ~12% increase of the module size.

Simpleton explanation?

 

[NostaSeronx]

Simpleton explanation?

CMT is not the reason Bulldozer/Piledriver/Steamroller/Excavator is phat @ 30.9 mm2(BD/PD) / 29.47 mm²(SR) / 22.5 mm²(XV). It is instead; the transistors of the non-critical type and very awful EDA tools.

With FBB and RBB, transistors do not need to have longer lengths. As longer lengths means less density.

Going with the point;
A 28FDSOI, but mostly a 22FDSOI Bulldozer-derived architecture can achieve a higher transistor density over 14nm FinFET Zen. Simply, because FBB and RBB can restrict the need of PBxx[Polybiasing].

 

[DrMrLordX]

Nosta, do you really think, at this point, that AMD will throw Zen and their adaptation of 14nm LPP into the trashcan in favor of XV on 22nm FDSOI?

Or do you really think that AMD will launch ANY XV product on ANY market based on XV using a 22nm SOI process?

I'm thinking "no" on both counts.

 

[NostaSeronx]

Nosta, do you really think, at this point, that AMD will throw Zen and their adaptation of 14nm LPP into the trashcan in favor of XV on 22nm FDSOI?

XV on 22FDSOI wouldn't support FP256.

That FP256 bit would need to be Value: 1 for that architecture to even be thought of releasing on 22FDX. Zen/K12 only support FP128, because AArch64 only supports up to FP128 media instructions.

Zen/17h development is at Sunnyvale.
Bulldozer/15h development is at Boston.

The development of two is asynchronous and the development of post-XV would have started in ultra late 2013.
---
If this was pre-2008 in alternate reality... (just for purposes of comparing that the whole question is silly)
[DrMrLordX, do you really think, at this point, that AMD will throw Bulldozer into the trashcan in favor of Stars on 45nm PDSOI?

Or do you really think that AMD will launch ANY Stars product on ANY market based on Stars using a 45nm SOI process?]

 

[TurtleCrusher]

CMT is not the reason Bulldozer/Piledriver/Steamroller/Excavator is phat @ 30.9 mm2(BD/PD) / 29.47 mm²(SR) / 22.5 mm²(XV). It is instead; the transistors of the non-critical type and very awful EDA tools.

With FBB and RBB, transistors do not need to have longer lengths. As longer lengths means less density.

Going with the point;
A 28FDSOI, but mostly a 22FDSOI Bulldozer-derived architecture can achieve a higher transistor density over 14nm FinFET Zen. Simply, because FBB and RBB can restrict the need of PBxx[Polybiasing].

Interesting stuff. Why didn't 22nm happen a long time ago?

It's probably been said before, but which socket would this be released on?

 

[DrMrLordX]

XV on 22FDSOI wouldn't support FP256.

. . . tangent much?

Zen/17h development is at Sunnyvale.
Bulldozer/15h development is at Boston.

Again, tangent much?

The development of two is asynchronous and the development of post-XV would have started in ultra late 2013.

Garbage. They have limited funds, and they are laying people off left and right. They are not going to undercut their new server darling with high-cost 22nm SOI chips, and NOTHING about 22nm SOI has shown up on their roadmaps. There was an announcement a few months ago that AMD was going to use a variant of the 22nm process they picked up from IBM for ultra-low power devices aimed at the custom embedded market. edit: 22FDX

If this was pre-2008 in alternate reality...

Complete and utter nonsense. AMD had no other irons in the fire except Stars back then. Nobody announced that Stars was dead.

Construction cores have been EOLed. Bristol Ridge is the last XV core to show up on any of AMD's roadmaps.

 

[DrMrLordX]

Interesting stuff. Why didn't 22nm happen a long time ago?

IBM's 22nm SOI node is expensive. It's what they use for POWER8. Presumably, AMD has had at least peripheral access to the process for awhile, though GF has made no obvious attempt to adapt their own facilities to manufacture on that node. The only reason why AMD has direct phsyical access to it now is that they picked up all of IBM's fabs, including those that crank out POWER chips.

 

[Dresdenboy]

Wait, 14/16nm Cat cores are still happening? Whaaat

AMD might need them for this kind of fun:

Ok, without further evidence, Zen might be the small core and Nosta's 256b FMA XV+ (would this be "Generation 2" ?) the big one, hehe.

 

[Arachnotronic]

AMD confirmed at its analyst day that the Cat cores are toast.

 

[Tuna-Fish]

As I said earlier in this thread, the cat cores will probably not be developed into new products, however they will continue to be manufactured for the current-gen console chips for at least half a decade. It will eventually make sense to shrink them, and I suspect that's what the cat-themed core projects are.

Also, since that is R&D that comes out of the vendor's pockets, not AMD, it's something that's not influenced by AMD's current financial woes.

However, the intended use as console chips also means that they really don't want to do any further improvements apart from things that cut power use. These are not interesting chips that might be useful in other applications.

 

[UserNameValid]

Honestly decided to post because it's just painful to read this thread.
Posting as someone who works for a partner company. NostaSeronx is talking out of his ass. Nobody from AMD even mentions bulldozer anymore.
As for Zen, does anyone wonder why there haven't been any leaks? It's because AMD doesn't say anything about it even to partners! All we know is the same thing as the post in OP. We're meeting all expectations, we've met all the targets, everything works great, blah blah blah, same thing every time. This is the first time that I can remember. Before this they were talking about K12 products all the time, how it's going to be leading performance and efficiency, best product, leadership, etc and they kept telling us we'd have it running by 2016 Q1. Then suddenly 2015 they pretty much stopped talking about it and then it's delayed to 2017. So they start saying that Zen products will be amazing and we're gonna have it running by 2016. And at this point all we know is that they're meeting expectations.
Now to me this looks like things aren't actually as great as they say. They used to boast their upcoming products all the time. And now nothing? I have a feeling Zen isn't as great as they claim.
As for release date, they said we'll know more at 2016 Q2. So at Q2 we're actually going to know when it's released(and they claim it will release the same year). They did say we're going to get actual numbers in 2016 Q1 so that's something. And this isn't confirmed but from what I've heard it that they have >10000 guaranteed sales from other partners already. Like what the hell we barely have any info. As far as I know everyone knows just as little. And everyone's already buying them? This makes no sense at all.
Also for anyone hoping to upgrade their desktops AMD says there's a chance that Zen products will be in low supply at first and if there's a high enough demand from partners, AMD is willing to push release date to 2017 for consumers. So now you don't have to wonder how it will do against Skylake, wonder about how it will do against CannonLake.

EDIT:

They did say we're going to get actual numbers in 2016 Q1 so that's something.

Just want to clarify. By actual numbers I mean specifications and possibly performance. They might announce this publicly. Even if they don't they are sending it out to partners so someone should leak it.

 

[Arachnotronic]

Honestly decided to post because it's just painful to read this thread.
Posting as someone who works for a partner company. NostaSeronx is talking out of his ass. Nobody from AMD even mentions bulldozer anymore.
As for Zen, does anyone wonder why there haven't been any leaks? It's because AMD doesn't say anything about it even to partners! All we know is the same thing as the post in OP. We're meeting all expectations, we've met all the targets, everything works great, blah blah blah, same thing every time. This is the first time that I can remember. Before this they were talking about K12 products all the time, how it's going to be leading performance and efficiency, best product, leadership, etc and they kept telling us we'd have it running by 2016 Q1. Then suddenly 2015 they pretty much stopped talking about it and then it's delayed to 2017. So they start saying that Zen products will be amazing and we're gonna have it running by 2016. And at this point all we know is that they're meeting expectations.
Now to me this looks like things aren't actually as great as they say. They used to boast their upcoming products all the time. And now nothing? I have a feeling Zen isn't as great as they claim.
As for release date, they said we'll know more at 2016 Q2. So at Q2 we're actually going to know when it's released(and they claim it will release the same year). They did say we're going to get actual numbers in 2016 Q1 so that's something. And this isn't confirmed but from what I've heard it that they have >10000 guaranteed sales from other partners already. Like what the hell we barely have any info. As far as I know everyone knows just as little. And everyone's already buying them? This makes no sense at all.
Also for anyone hoping to upgrade their desktops AMD says there's a chance that Zen products will be in low supply at first and if there's a high enough demand from partners, AMD is willing to push release date to 2017 for consumers. So now you don't have to wonder how it will do against Skylake, wonder about how it will do against CannonLake.

Easily the most useful post in this thread. Thanks for the color on the situation.

 

[frozentundra123456]

Exactly. A dose of reality.

 

[Arachnotronic]

Exactly. A dose of reality.

Honestly it sounds to me that AMD isn't hitting its frequency targets with Zen, which wouldn't be a surprise because doing high IPC + high frequency is HARD.

 

[coercitiv]

Easily the most useful post in this thread.

New user with no (established) history writes first post to complain about tech company not revealing product info, then comments on rumors of guaranteed sales and expresses concern regarding product performance. Based on no product info.

Veteran user with vast experience on many product launches, including from said tech company, comes in and labels the post as "most useful post in this thread". Based on no product info.

 

[frozentundra123456]

Well, it is "useful" just to slow the hype train a bit. Zen may in fact be great, but the speculation has gotten way, way out of hand.

 

[UserNameValid]

New user with no (established) history writes first post to complain about tech company not revealing product info, then comments on rumors of guaranteed sales and expresses concern regarding product performance. Based on no product info.

Veteran user with vast experience on many product launches, including from said tech company, comes in and labels the post as "most useful post in this thread". Based on no product info.

I wouldn't mind posting with my real account that has more posts, but if you googled my username, you would find my real name. So I made a new account and posting via proxy.
The reason I'm concerned about product performance is because AMD usually provides a lot more information to partner companies. Maybe I am jumping to conclusions a bit too early. They did change CEO recently so maybe that has something to do with it. But they are being VERY secretive with Zen products.

EDIT:
@Dresdenboy
I can't send PMs without 25 posts.

Multiple accounts are not allowed.
Markfw900

 

[sm625]

I dont care what they say as long as they make a processor that can score 50k on google octane. I seriously doubt the first one will even get above 40k. Cannonlake looks to be targetting 50k so that is what AMD needs to be targetting.

 

[raghu78]

With AMD its better to have low expectations. I expect they will disappoint with performance. AMD has become all talk and not much about execution. Nvidia and Intel are hammering AMD to its imminent death. As an underdog supporter I still hope AMD can bring out competitive products in 2016/2017 and recover towards long term sustainable profitability.