Fudzilla: New AMD Zen APU boasts up to 16 cores (plus Greenland GPU with HBM)

[DrMrLordX]

If AMD has the resources to launch Zen-based APUs in Q3 2016, then I have to wonder where that leaves Bristol Ridge? Why launch a 28nm Excavator quad APU when you have the tech to do a Zen quad APU at what will probably be a lower cost (14nm vs 28nm) with fewer yield issues (smaller die)? I'm a little confused on that.

 

[Shehriazad]

If AMD has the resources to launch Zen-based APUs in Q3 2016, then I have to wonder where that leaves Bristol Ridge? Why launch a 28nm Excavator quad APU when you have the tech to do a Zen quad APU at what will probably be a lower cost (14nm vs 28nm) with fewer yield issues (smaller die)? I'm a little confused on that.

Eh? Wouldn't those 28nm be DIRT cheap to produce? Low end people like to buy shiny things, too.

 

[erunion]

If AMD has the resources to launch Zen-based APUs in Q3 2016, then I have to wonder where that leaves Bristol Ridge? Why launch a 28nm Excavator quad APU when you have the tech to do a Zen quad APU at what will probably be a lower cost (14nm vs 28nm) with fewer yield issues (smaller die)? I'm a little confused on that.

AMD hasn't explain that yet as far as I know. Their roadmap shows both a big APU( kaveri successor) and a little APU(Beema successor). Maybe Zen will appear on the little APU first and the big APU will stay excavator for a bit longer?

 

[DrMrLordX]

Eh? Wouldn't those 28nm be DIRT cheap to produce? Low end people like to buy shiny things, too.

Eh depends on the cost per transistor. In theory, 14nm should be a lot cheaper, assuming they don't have massive yield problems, tool prices aren't insanely high, etc.

Yeah 28nm is established, the fabs are pretty-much paid for already, and so forth and so on, but assuming similar transistor count per die, you could squeeze out a lot more dice per wafer using 14nm vs 28nm.

Then there's the good old WSA. If GF is going to be cranking out 28nm wafers and by Jove somebody'd better use em, then AMD may be on the hook to use those wafers for something. Nolan is going to be/supposed to be on somebody's 20nm process, so it's out. They sure as heck aren't going to want to keep cranking out Jaguar-based chips unless they really have to (and by then, they should have Puma retired as well), and with FM2+ being EoLed in 2016, I guess Bristol Ridge on FM3 may just be a sink for "we bought em 'cuz they made em" wafers from the 28nm fabs that GF does not plan to idle by that point.

AMD hasn't explain that yet as far as I know. Their roadmap shows both a big APU( kaveri successor) and a little APU(Beema successor). Maybe Zen will appear on the little APU first and the big APU will stay excavator for a bit longer?

We know the Mullins/Beema successor is supposed to be Nolan (Amur on the ARM side), and various software providers have dropped hints that they are already supporting it (AIDA64, for example). So Nolan is probably coming soon-ish, and it's Puma+. Everything else is murky. First we had leaks that Zen would be 8c/16t @ 95w, now we have word that it could go up to 32c/64t, and news of this 200W-300W APU is also quite . . . interesting.

 

[cytg111]

This is not meant to be a flame post...I'm just saying...people seem to over/underestimate wattage all the time. And mind you...that Intel Chip is not stuffed with a fat GPU.

AMD is maintaining two lines here right? Big cores -> I dont give a crap about wattage and smaller, cat, cores that is targetted perf/watt. They're not trying to pull a ie. haswell 4.5w -> 95w.

 

[erunion]

AMD is maintaining two lines here right? Big cores -> I dont give a crap about wattage and smaller, cat, cores that is targetted perf/watt. They're not trying to pull a ie. haswell 4.5w -> 95w.

They are not. Zen is the only x86 core in development, the future is Zen.

 

[cytg111]

They are not. Zen is the only x86 core in development, the future is Zen.

So the message is that they're getting out of mobile x86 ?

 

[Yuriman]

So the message is that they're getting out of mobile x86 ?

Unless Zen is a small core. In which case, they're getting out of high performance desktop.

 

[erunion]

Personally, I speculate AMD will replace the little APUs with 4 Zen cores. And the big APUs with 8 Zen cores.

I also think the 300W "APU" will be a high end GPU with a few CPU cores to make it bootable and live in the socket. Intel has Knight Landing and Nvidia is adding ARM cores to their GPUs, so that will be AMDs answer.

 

[antihelten]

Eh....wouldn't be too sure about that.

You go from 2 modules to 4 actual cores (most likely with extra threads).

You go from 8 GPUs to 12.

You add HBM.


Since AMD also wants to walk more toward SoC style...that is possible.

Also the reports say somewhere from 200W - 300W. AMD wouldn't possibly be able to already give us any sort of real value since I'm pretty sure that there won't even be an actual working sample out, yet. (Because someone would've leaked the heck out of that, already xD)


Let's say they reach 250W...perfectly adds up. I can see the new cores eating up 100ish Watt on their own as they would be more close to a FX8XXX series (4 ACTUAL cores, 8 threads) instead of what current APUs have. Then AMDs GPUs...if it's anywhere near a X70 or X80 model like I suggested...that can easily sneak in another 100W+ Watts...we know that AMD isn't too great with their GPU wattage.
Then you have another 50ish for whatever else they may add + some HBM. (which of course won't be sucking that much).

And we don't even know what those cores clock at...AMD likes to impress with high clocks rather than high IPC. Even with the new architecture I can already see AMD forcing their top end chips to the 4+ Ghz area by default.


But yea...SOMEWHERE BETWEEN 200-300 is what that "news article" said...not 300. For all we know it could end up being 210w.


Edit:I don't see how my prediction is unrealistic...see what happens to Intels new chip when pushing it just to 4 Ghz. AMD loves running their stuff at 4 Ghz base on their top chips.

This is not meant to be a flame post...I'm just saying...people seem to over/underestimate wattage all the time. And mind you...that Intel Chip is not stuffed with a fat GPU.

12 GPU cores would be 768 shaders, aka an R7 260 (or something like an R7 350 if it's rebranded), in other words nowhere near an X70/X80 model.

The R7 260 has a TDP of 95W, which should of course drop a fair bit with a new process. So to hit a total TDP of 200-300W, the 4 CPU cores would have to account for 150-200W. This would mean that Zen is even more power hungry than a 4 module Bulldozer CPU even on a newer process, which seems highly unlikely.

Also your power graph lists total system power, not CPU power.

 

[Shehriazad]

12 GPU cores would be 768 shaders, aka an R7 260 (or something like an R7 350 if it's rebranded), in other words nowhere near an X70/X80 model.

The R7 260 has a TDP of 95W, which should of course drop a fair bit with a new process. So to hit a total TDP of 200-300W, the 4 CPU cores would have to account for 150-200W. This would mean that Zen is even more power hungry than a 4 module Bulldozer CPU even on a newer process, which seems highly unlikely.

Also your power graph lists total system power, not CPU power.

Your metric might not apply to the new architecture. We can't possibly know anything about the GPU cores inside of that chip. Especially since we are talking about Zen supposedly featuring R400 series GPU cores...and 300 isn't even out yet. We might even talk about GPU core clusters instead of "cores". AMD likes clustering/modularing things together...I can almost see it happening xD

And again...people still get way too hung up about the watts...and As you can see even Intels new big core 8 core CPU that comes with a base "tdp" of 140W at 3 Ghz easily puts on another 60watts for getting it to 4ghz. And again...those 200-300 were a random estimate. Chip might release at 300...250..200...or just 150. AMD doesn't even know themselves yet, I'm pretty sure...while they are probably having a grasp on Zen now, should it seriously happen, there is just no way they actually already got it up and running with new cores, new R400 series, HBM, and DDR4 on 14nm...not with those R&D funds


AMD said Zen is the end of their Modular CPU design...anyone assuming that the chip would be a 16 core CPU and then ALSO fit HBM, Quad Channel DDR4 AND a big GPU must be a bit insane in my eyes.

There would be no physical space for ALL of that unless you make the chip physically huge...which would drive up the prices insanely if I'm not mistaken.

Not to mention...the market for 16 core CPU users seems rather small in the standard Desktop market. And now don't yell DX12 DX12...that might help for games...but until games will properly benefit from more than 8 cores...that's gonna be a whiiiiiile.

 

[DrMrLordX]

So the message is that they're getting out of mobile x86 ?

Probably not. Take a look at what they did shifting from Kaveri to Carrizo: they went from a ~95W TDP chip with a few "mobile-ish" features (cTDP throttling, grr) that only really worked on the desktop to a max 35W TDP chip that is definitely going into mobile products. And they did that on the same basic process tech (yeah, they incorporated HDL, but still . . .)

The alleged base Zen part is 8c/16t with a TDP of 95W on a 14nm process. You don't think they could pull off a 2c/4t part for mobile, or maybe even 4c/8t? I think they have plans to use the Zen cores in just about everything, except maybe the tablet space. I expect that Nolan/Amur will be their tablet chip for awhile.

12 GPU cores would be 768 shaders, aka an R7 260 (or something like an R7 350 if it's rebranded), in other words nowhere near an X70/X80 model.

The R7 260 has a TDP of 95W, which should of course drop a fair bit with a new process. So to hit a total TDP of 200-300W, the 4 CPU cores would have to account for 150-200W. This would mean that Zen is even more power hungry than a 4 module Bulldozer CPU even on a newer process, which seems highly unlikely.

There would probably be more than 768 shaders.

 

[TurtleCrusher]

200-300w on 14nm? It'd seriously have to be a complete monster of a CPU. If Zen is focused on efficiency and then cranked out a high clocking chip... Exciting, interesting, but I'm left wanting more information before I start making too many assumptions.

I'm optimistic that they can pull it off and I really hope as a consumer and enthusiast that it is the case, but the odds aren't in their favor money-wise. I hope this is K8/K8x2 all over again but with graphics as well. The better AMD does with Zen, it would result in lower prices with higher performance across the market. That is something I think we all can agree on. (Besides all the shills and investors)

 

[TurtleCrusher]

And on the topic of die space, if HBM is addressable to the CPU, I'd consider it an L4 cache and the L3 would be 2MB or less, similar to original phenom. It'd also make a push for higher speed/lower capacity cache knowing there's a ton of addressable space in HBM.

 

[ARMD Supremacy]

Unless Zen is a small core. In which case, they're getting out of high performance desktop.

ZDNet's story is actually about an Apple chip deal, but it names AMD as one of Samsung's manufacturing clients:

Samsung is expected to start producing application processors (APs) for clients such as Apple, Qualcomm, and AMD, using its 14-nanometre process around the end of the year.

Report was from late 2014 and it indicates AMD probably has number of surprises in store. Whether it is semi-custom APUs for Samsung or other companies is yet to be disclosed, but it is certain that they are on path to have very competitive x86 and ARM platforms in 14nm process, sooner rather than later.

http://techreport.com/news/27149/report-samsung-to-produce-14-nm-chips-for-amd

 

[Idontcare]

Report - Samsung to produce 14nm chips for AMD

You know this is the very kind of BS posting that gave AMD the black-eye it got from the "it goes to 11" hype that JFAMD gave Bulldozer.

You aren't an idiot. You can read the very link you posted on, and if you click on the source and read it then you find just one, and only one, reference to AMD which reads as follows:

Samsung is expected to start producing application processors (APs) for clients such as Apple, Qualcomm, and AMD, using its 14-nanometre process around the end of the year.

There isn't a single other reference to AMD in the entirety of the "article" and yet here you are, parroting the crap as if it were gospel fact.

Trash your reputation around here if that suits you, but for your own sake click a few times into the sources of your own links before you get all gobs excited about posting a new link to it here.

Folks here are a wee bit more intelligent about drilling down through the BS to find out who or what is actually doing the reporting, and in the end you will just come out looking the fool with these sensationalism postings.

(or you don't care that this is the impression you are leaving, in which case who cares about my particular take on them?)

 

[ARMD Supremacy]

There isn't a single other reference to AMD in the entirety of the "article" and yet here you are, parroting the crap as if it were gospel fact.

Trash your reputation around here if that suits you, but for your own sake click a few times into the sources of your own links before you get all gobs excited about posting a new link to it here.

Folks here are a wee bit more intelligent about drilling down through the BS to find out who or what is actually doing the reporting, and in the end you will just come out looking the fool with these sensationalism postings.

(or you don't care that this is the impression you are leaving, in which case who cares about my particular take on them?)

The original report came from Zdnet Korea from a Korean journalist Cho Mu-hyun who covered the conference by Samsung's LSI president.

Kim Ki-nam, president of the Korean electronic giant's semiconductor business and head of System LSI business, told reporters at Samsung's headquarters in Seoul that once the company begins to supply Apple with chips using its latest technology, profits "will improve positively".

Samsung is expected to start producing application processors (APs) for clients such as Apple, Qualcomm, and AMD, using its 14-nanometre process around the end of the year.

Kim declined to comment on when Samsung will start mass producing said chips for clients.

http://www.zdnet.com/article/samsung-expects-sales-boost-from-apple-chip-deal/

The fact is that the company name AMD came from Zdnet Korea and from a reporter in Korea covering Samsung, so it is more than BS from out of no where as you state.

 

[mrmt]

The fact is that the company name AMD came from Zdnet Korea and from a reporter in Korea covering Samsung, so it is more than BS from out of no where as you state.

AMD would have a lot of explaining to do because of the WSA remarks the management team made in the last two years. As of *now* they can't manufacture CPUs outside GLF, and it would be a real snafu if they have negotiated something with Samsung and GLF to allow the former to manufacture chips for them and did not brief their investors of the results of this negotiation.

So unless this south korean journalist is very, very good and caught AMD management with the pants down, this is news is BS.

 

[Abwx]

The original report came from Zdnet Korea from a Korean journalist Cho Mu-hyun who covered the conference by Samsung's LSI president.

The fact is that the company name AMD came from Zdnet Korea and from a reporter in Korea covering Samsung, so it is more than BS from out of no where as you state.

In the current context this news make perfect sense.

Since AMD wants to release Zen in 2016 they need to have functional silicon one year at least before launch, this put the target at Q2/Q3 2015 for the first ES and it s unlikely that GF can supply thoses in this timeline.

Luckily for AMD they use Samsung process, so AMD can manufacture their ES at Samsung for the validation phase without the hassle to repeat again this work once they manufacture the chips at GF.

 

[Enigmoid]

And on the topic of die space, if HBM is addressable to the CPU, I'd consider it an L4 cache and the L3 would be 2MB or less, similar to original phenom. It'd also make a push for higher speed/lower capacity cache knowing there's a ton of addressable space in HBM.

2 MB L3 on a 16 core chip is useless. Given AMD's poor L3 cache performance I think they would just ditch it.

 

[Vesku]

You know this is the very kind of BS posting that gave AMD the black-eye it got from the "it goes to 11" hype that JFAMD gave Bulldozer.

You aren't an idiot. You can read the very link you posted on, and if you click on the source and read it then you find just one, and only one, reference to AMD which reads as follows:



There isn't a single other reference to AMD in the entirety of the "article" and yet here you are, parroting the crap as if it were gospel fact.

Trash your reputation around here if that suits you, but for your own sake click a few times into the sources of your own links before you get all gobs excited about posting a new link to it here.

Folks here are a wee bit more intelligent about drilling down through the BS to find out who or what is actually doing the reporting, and in the end you will just come out looking the fool with these sensationalism postings.

(or you don't care that this is the impression you are leaving, in which case who cares about my particular take on them?)

Yes, given AMD's track record I'm currently expecting them to have retail 14nm something like 6-9 months after the non-Intel 14nm mobile chips show up. Up to AMD to deliver, it would be very foolish to make any purchase planning around their "expected" release cycles.

 

[Fox5]

Probably not. Take a look at what they did shifting from Kaveri to Carrizo: they went from a ~95W TDP chip with a few "mobile-ish" features (cTDP throttling, grr) that only really worked on the desktop to a max 35W TDP chip that is definitely going into mobile products. And they did that on the same basic process tech (yeah, they incorporated HDL, but still . . .)

The alleged base Zen part is 8c/16t with a TDP of 95W on a 14nm process. You don't think they could pull off a 2c/4t part for mobile, or maybe even 4c/8t? I think they have plans to use the Zen cores in just about everything, except maybe the tablet space. I expect that Nolan/Amur will be their tablet chip for awhile.



There would probably be more than 768 shaders.

The current processes AMD is using are probably better optimized for low power than for high performance. Look for cpu clocks to be down, and cores to be up.

200-300w on 14nm? It'd seriously have to be a complete monster of a CPU. If Zen is focused on efficiency and then cranked out a high clocking chip... Exciting, interesting, but I'm left wanting more information before I start making too many assumptions.

Look for something similar to the console APUs. Tiny CPU cores (probably taking up like 50W of the power budget at most) and gigantic GPU.

 

[frozentundra123456]

I dont know, I just dont see the market for that. AMD has had a better gpu than intel for years and they have steadily lost market share. Now granted, if HBM allows the igpu to really shine compared to current igps, there will be a market for it. Question is how big is that market, and they will also be competing with themselves (vs both consoles and dgpus).
Plus dgpus will apparently have HBM before Zen does, and with that and 14nm dgpus eventually, plus higher resolutions becoming mainstream, even with great improvements, APUs are shooting at a continually moving target.

My feeling is that the 16 core rumor either refers to a server only chip, or to "compute cores", meaning cpu and gpu cores counted together. Or the rumor could just be complete BS.

 

[AtenRa]

As I have said before, if that chip ever been manufactured at 14nm, it will only be for the HPC/Server market. The Consumer TAM for such a Chip is negligible for AMD to even think of releasing it.

 

[artivix]

In the current context this news make perfect sense.

Since AMD wants to release Zen in 2016 they need to have functional silicon one year at least before launch, this put the target at Q2/Q3 2015 for the first ES and it s unlikely that GF can supply thoses in this timeline.

Luckily for AMD they use Samsung process, so AMD can manufacture their ES at Samsung for the validation phase without the hassle to repeat again this work once they manufacture the chips at GF.

You look to be right about it because a search on the news releases on the Samsung website also has AMD exec Lisa Su indicating on their benefits from the Samsung GF tech partnership on advanced processes...

Through this multi-year exclusive technology license, process design kits (PDKs) are available now, allowing customers to start designing with models, design rule manuals, and technology files that have been developed based on silicon results from 14nm FinFET test chips. Mass production for the 14nm FinFET technology will begin at the end of 2014.

"This unprecedented collaboration will result in a global capacity footprint for 14nm FinFET technology that provides AMD with enhanced capabilities to bring our innovative IP into silicon on leading-edge technologies," said Lisa Su, senior vice president and general manager of Global Business Units at AMD.

"The work that GLOBALFOUNDRIES and Samsung are doing together will help AMD deliver our next generation of groundbreaking products with new levels of processing and graphics capabilities to devices ranging from low-power mobile devices, to next-generation dense servers to high-performance embedded solutions."

http://www.samsung.com/global/business/semiconductor/news-events/press-releases/detail?newsId=13364