[AMD] K12 will be on 28nm

[jdubs03]

This is a pretty poor choice for process node especially considering that competitor solutions will be laid out on 14nm/16nmFF from Samsung, TSMC, and of course Intel. And by that time Intel will likely be preparing for 10nm in the not too distant future from K12 products being released.

I'm sure that the architecture will likely be competitive on a relative basis with their competitors, but to target the top-performing SoC's with a 2 to 2.5 node disadvantage is impossible. And that's sad because, they would have a shot at competitiveness on the same node as say Nvidia and Qualcomm.

Cost may be a factor, but what benefit is there when all of their competitors destroy AMD on performance and power consumption? I just can't see the utility in this choice, I didn't even like the Tegra K1 being on 28nm (though that's a pretty damn good solution, plus it leaves the M1 and V1 for 20nm and then 16nmFF+ in 2015 and 2016). If AMD doesn't at least go to 20nm, they won't be able to get any sales growth and market share gains, and I forsee them getting purchased by 2020.

 

[frozentundra123456]

I dont know what to think. I really am not sure this statement is correct. After all, it is very vague and worded extremely poorly. I cant believe AMD would really try to bring out a new architecture almost 2 years into the future on 28 nm. If true though, it is a disaster, but perhaps he just was confused and misspoke.

 

[Idontcare]

This is all part and parcel to AMD continuing the process of morphing into the next Via. Neither bankrupt nor relevant.

 

[cbn]

28nm will be bad for mobile and probably server.

But for performance desktop I doubt it would be much of a detriment if AMD focused on improving IPC and got rid of a lot of the extra iGPU on current chips. (but then again I don't know how big of a market enthusiast desktop really is?)

 

[ThatBuzzkiller]

They may as well just cancel the project altogether at this point ...

 

[mrmt]

"some 20nm" product in 2015. sort of strategically ambiguous on when they introduce a product with finfet.

He probably was talking about high end GPUs, they can't really afford to stay on 28nm against Nvidia.

i like how he plays down the importance of process node.

That's a sad spectacle in a conference like that. He knows that a node is important, he knows that not having a bleeding edge node will have consequence for its current business model, his audience also knows that, but since AMD isn't forthcoming enough with its future strategy, he has to tap dance around the issue.

Got to be a mistake. If that is true, it's utterly DOA.

EDIT: A link to the source would be appreciated, thanks!

Check for yourself, I have included a link for the transcript and for the audio itself. Yes, it is DOA on the consumer market.

In light of this news I think it's more likely that K12 and Zen will be large cores.

Really doubt it. They need a descent core count on servers, and they wouldn't be able to reach with with a way too big core.

 

[NTMBK]

Check for yourself, I have included a link for the transcript and for the audio itself. Yes, it is DOA on the consumer market.

Thanks, appreciated. Yeah, definitely sounds like he's saying 28nm. *facepalm*

 

[Idontcare]

That's a sad spectacle in a conference like that. He knows that a node is important, he knows that not having a bleeding edge node will have consequence for its current business model, his audience also knows that, but since AMD isn't forthcoming enough with its future strategy, he has to tap dance around the issue.

This is the bottom line.

Only the most naive member of the audience wouldn't appreciate the gravity or reality of this point.

I'm sure he personally is enjoying the paycheck being direct-deposited every other week but at the same time I would be very surprised if he isn't planning his "what's next for me" phase of his career path because he is in the position to know that there are only so many death knells one can overlook in their present employer's future before pulling the ripcord and bailing.

Rory has only been out for a manner of days yet, we ignorant laymen in the public domain have no idea the shoes that are yet to drop. But he does, and his dancing is all the more insightful because of that immutable fact.

 

[ShintaiDK]

28nm will be bad for mobile and probably server.

But for performance desktop I doubt it would be much of a detriment if AMD focused on improving IPC and got rid of a lot of the extra iGPU on current chips. (but then again I don't know how big of a market enthusiast desktop really is?)

Not going to happen. No cash for it, no node for it. And ARM is getting all the focus from AMD.

 

[ShintaiDK]

They may as well just cancel the project altogether at this point ...

It still gets money back one way or the other. Just look at VIA.

But AMD is gone in anything performance related on the CPU front.

Its sad to think on that AMD bought ATI, only to be a new ATI.

 

[krumme]

This is the bottom line.

Only the most naive member of the audience wouldn't appreciate the gravity or reality of this point.

I'm sure he personally is enjoying the paycheck being direct-deposited every other week but at the same time I would be very surprised if he isn't planning his "what's next for me" phase of his career path because he is in the position to know that there are only so many death knells one can overlook in their present employer's future before pulling the ripcord and bailing.

Rory has only been out for a manner of days yet, we ignorant laymen in the public domain have no idea the shoes that are yet to drop. But he does, and his dancing is all the more insightful because of that immutable fact.

Your personal remarks about a CFO - paychecks/whats next whatever - is not worthy.

Does it cross your mind he actually have his reasons?

AMD is tied to GF by a WSA. Kumar is working to get the relationsship to work, dispite GF not delivering anything for years. His public speaking have to reflect that.
At the same time there is constant negotiation between the stakeholders - AMD pressing for better process for the $ - flat out "buying"- so to speak process for 2016 beforehand in public is not in AMD interest as it could inflate the price/terms for it.

The need for constant negotiating the terms of the wsa is a huge mess, but that mess is very much out of the hands of the CFO - but in the hands of Mubadala ownership. He does the only thing he can; adressing the positive momentum in the relationsship. That is the only good thing to do from a management perspective.

Yes K12 on 28nm is obviously a dud and will relocate it to niche, niche market if it happens. But thats very different from "paychecks" and the constant trolling post from mrmt about eg. amateurish.

 

[AtenRa]

Intel SoFIA is a 28nm all the way to the end of 2015. It targets the Entry/Value and ULC market.

http://forwardthinking.pcmag.com/none/329777-intel-updates-roadmaps-for-2015

K12(x86) and Zen(ARM) at 28nm SKUs, could be the AMD equivalent for entry/value/ULC Tablets/AiO/mobile etc.

This is the only explanation i can come with for k12/zen to be at 28nm.
AMD have to clear this up.

 

[krumme]

Intel SoFIA is a 28nm all the way to the end of 2015. It targets the Entry/Value and ULC market.

http://forwardthinking.pcmag.com/none/329777-intel-updates-roadmaps-for-2015

K12(x86) and Zen(ARM) at 28nm SKUs, could be the AMD equivalent for entry/value/ULC Tablets/AiO/mobile etc.

This is the only explanation i can come with for k12/zen to be at 28nm.
AMD have to clear this up.

Think process at GF, and Mubadala strategy, as much as products and markets.
If 16nm finfet is going to get expensive at GF (low yield) can K12/zen pay for it?

I have my doubt about that, as GF just got their 28nm working. Its not made for high-end phones (lack everything except the apu part), and the value servermarket is more than bleak as high ipc get huge advantages here. And if GF even had 16nm working fine, would Mubadala not prioritize other customers not tied to GF?

That leaves value end. Sucks as enthusiast, but at least its probably not huge and power consuming as BD cores. But if SS puls the process ahead and GF follows and will introduce it at 16nm fitfet then just great.

 

[Arachnotronic]

Intel SoFIA is a 28nm all the way to the end of 2015. It targets the Entry/Value and ULC market.

http://forwardthinking.pcmag.com/none/329777-intel-updates-roadmaps-for-2015

K12(x86) and Zen(ARM) at 28nm SKUs, could be the AMD equivalent for entry/value/ULC Tablets/AiO/mobile etc.

This is the only explanation i can come with for k12/zen to be at 28nm.
AMD have to clear this up.

SoFIA is a cheap smartphone chip with an integrated modem. Last I checked, AMD wasn't in the mobile market.

 

[mrmt]

Y
Yes K12 on 28nm is obviously a dud and will relocate it to niche, niche market if it happens. But thats very different from "paychecks" and the constant trolling post from mrmt about eg. amateurish.

Do you think AMD management is the baseline for professionalism?

SoFIA is a cheap smartphone chip with an integrated modem. Last I checked, AMD wasn't in the mobile market.

AMD has 0 radio IP, so they are a no-go for the mobile market.

 

[Headfoot]

the only way that could not suck is if 28nm FDSOI is the 28nm process they're talking about, FDSOI actually somehow lives up to the hype, and is also on time. So basically, it's the longest of long shots. It's probably just a value chipset on el-cheapo bulk 28nm in 2016 for bargain basement applications: e.g. AMD begins competing against VIA...

AMD's only CPU shot now is to turn the SeaMicro IP and Seattle into a continually iterating strategy and actually lead in the ARM 64-bit microserver/scale-out space. If they can capture an appreciable portion of this growing marketspace they might finally see some real growth. I'll believe it when I see it though. That space is going to be very competitive. If 2016 is the first time they update Seattle and it's with K12 cores at 28nm, then they're done.

 

[blake0812]

So i'm guessing 28nm is good? My i5 is 22nm, so if I had to guess, the smaller the processor the better?

 

[mrmt]

Rory has only been out for a manner of days yet, we ignorant laymen in the public domain have no idea the shoes that are yet to drop. But he does, and his dancing is all the more insightful because of that immutable fact.

We can now see why Rory Read abandoned his "predator", "aggressor", "attack and protect" vocabulary after a few quarters leading AMD. How would he compete against 10nm Intel chips with 28nm, let alone predate Intel?

We can also see how much clout GLF has over AMD. GLF will rather kill AMD than allow it to manufacture CPUs with other foundries. AMD will be handicapped even against other ARM IHVs in terms of foundry nodes. It won't be pretty.

It takes a lot of effort to thoroughly destroy a company the way Hector Ruiz did with AMD, the guy really deservers a space on management books.

 

[Phynaz]

This is all part and parcel to AMD continuing the process of morphing into the next Via. Neither bankrupt nor relevant.

You nailed it.

And I still blame Hector.

 

[formulav8]

Yeah Hector Ruinz. He was awful for AMD.

 

[NTMBK]

So i'm guessing 28nm is good? My i5 is 22nm, so if I had to guess, the smaller the processor the better?

The smaller the transistor, the better. Smaller transistor = more transistors on a chip = more circuits = faster processor, generally speaking. Also smaller transistor = lower power consumption (why Ivy Bridge is more efficient than Sandy Bridge).

 

[geoxile]

the only way that could not suck is if 28nm FDSOI is the 28nm process they're talking about, FDSOI actually somehow lives up to the hype, and is also on time. So basically, it's the longest of long shots. It's probably just a value chipset on el-cheapo bulk 28nm in 2016 for bargain basement applications: e.g. AMD begins competing against VIA...

AMD's only CPU shot now is to turn the SeaMicro IP and Seattle into a continually iterating strategy and actually lead in the ARM 64-bit microserver/scale-out space. If they can capture an appreciable portion of this growing marketspace they might finally see some real growth. I'll believe it when I see it though. That space is going to be very competitive. If 2016 is the first time they update Seattle and it's with K12 cores at 28nm, then they're done.

Pretty sure FD-SOI is ready now. And Advanced FD-SOI on 28nm is said to be comparable or superior to 20nm IIRC. So AMD would be a bit behind but not too far off if they used it in 2016.

Speaking of, it strikes me as strange that AMD is going 28nm. I doubt 16nm Finfet will be ready for large chips, so I suspect that Zen and K12 will be large cores on a simpler process than cutting edge.

 

[mrmt]

Pretty sure FD-SOI is ready now. And Advanced FD-SOI on 28nm is said to be comparable or superior to 20nm IIRC. So AMD would be a bit behind but not too far off if they used it in 2016.

You should tell that to IBM, that despite having access to 28nm FDSOI they went straight for a 22nm SOI process by themselves.

STM did a very good marketing job here, despite failing miserabily to generate sales of their 28nm FDSOI node, they still get a lot of press coverage of what they are supposed to deliver.

 

[krumme]

Do you think AMD management is the baseline for professionalism?

What is that for a strawman?
I dont beliewe there is any company the size of 10k, with so much highly qualified and educated personal as eg AMD, that would not have a professional CFO - or any of the others top positions, - and down the ranks.
The people at that level is perhaps eg. greedy or agressive - but they are never, ever amateurs at their job.
Its naive to think otherwise, and its only a showcase of own incompetence calling them for amateurs.
As for Kumar, ofcource he should have frased it differently, but give it a rest. The man does - and for one of the first time in years - the nessesary to motivate the relationship to the rest of the organization. That is proactive management.
Whining and crying for something (the wsa), you cant do anything about, is just bad management - and he does what should actually have been done 3 years ago - something to motivate and develop the relationship. No matter how bad GF process was or is, or how bad AMD BD design was. Thats irrelevant.
Thats professional management what he does - in a situation when its most difficult. For Mubadala AMD only exist as a mean to develop GF - as it was here it was thought, the huge potential was. Kumar doesnt concentrate on that - he concentrate on the relationsship to GF.
Talking business opportunities under those circumstances instead of crying of bad weather - Thats professional management imo. And thats the management we all want.

 

[MisterMac]

This is all part and parcel to AMD continuing the process of morphing into the next Via. Neither bankrupt nor relevant.

Ouch, man, Ouch.


And your probably the person in this forum - i'd value most serious in terms of "predicting microprocessor future".

I geuss if you see it like this - this is why we'll see more and more bloomberg\similar mainstream business sites now beginning to pick up doom and gloom articles.