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

[AtenRa]

He's talking about the Zen based FX series.

It has no iGPU.

That doesnt mean they abandoning the APUs.

 

[AtenRa]

Lets say hypothetically that Samsung 14nm also clocks lower than 32nm.
will you same thing about 14nm?

14nm FF is not a regression over 32nm SOI or even 20nm SOI and certainly not over a 28nm Planar.

The only thing 28nm brought over 32nm SOI was 10% higher density (without HDL) and lower cost (Planar vs SOI) at higher volumes. Electrical characteristics were inferior to 32nm SOI and thus the regression in clocks (for the same mArchtectural design).

 

[AtenRa]

I dont fault them for going out swinging, but I definitely question the way they are doing it.
They are going away from their area of expertise (APUs) just when HBM finally had a chance to really make the igp powerful enough to be a compelling purchase decision. And what about HSA? Why would a developer commit resources to HSA when AMD's only chip on their touted new architecture has no igpu? I thought even in servers HSA and microservers was supposed to be the magic sauce that would give them back marketshare. A major tenet of business planning is to concentrate on your own strengths and attack your competitor's weakest point. This is just the opposite, abandoning your greatest area of expertise (one the you paid 5 billion dollars for) and attacking the giant at their strongest point. This roadmap really does show that those who said AMD would be limited by their small research budget were in fact correct and not just picking on poor AMD.

You dont need ZEN to have HSA, Carrizo with Excavator cores will be the first fully enabled HSA APU in the world and in products in Q2 2015. Every new Generation of APUs from now on will be HSA, like Gen 7 next year.
Also, HBM is a new tech and more expensive than DDR-3/DDR-4. And since AMD still has the lead in the APU performance both in Desktop and Mobile (in the low/mid price segments) they dont need to rush to release an HBM APU.

 

[R0H1T]

I, for one, dont expect AMD to reveal anything.

But I disagree with the handful of people making grand claims in the absence of information.

Grand claims like Broadwell will be (insert xyz here) yeah I recall Intel parroting that line along with a number of posters on this forum or how Broadwell -> Skylake was to be a bigger leap than P4 -> Core

 

[CHADBOGA]

Grand claims like Broadwell will be (insert xyz here) yeah I recall Intel parroting that line along with a number of posters on this forum or how Broadwell -> Skylake was to be a bigger leap than P4 -> Core

You are deeply confused.

People may have referenced the original WCCF Tech article stating that, but find me someone who actually believed that.

Here is the thread, and not a single believer, yet you falsely claim people on this forum were peddling the new Conroe lie.

http://forums.anandtech.com/showthread.php?t=2425928

 

[jj109]

14nm FF is not a regression over 32nm SOI or even 20nm SOI and certainly not over a 28nm Planar.

For cellphones and tablets, sure. Can you say the same thing about high power CPUs? I don't know where this confidence is coming from. How can you seriously use Intel as an example of clocks getting lower on finfet mobile oriented nodes, and then be convinced that AMD is immune to the same effect?

Also, HBM is a new tech and more expensive than DDR-3/DDR-4. And since AMD still has the lead in the APU performance both in Desktop and Mobile (in the low/mid price segments) they dont need to rush to release an HBM APU.

Mid price mobile? 15W Broadwell i5s have caught up to 19W Kaveri in graphics (Tomb Raider, Bioshock, not 3DMark) and pull farther ahead in CPU performance. There's a gap between CT/Core that AMD can currently squeeze into, but if AMD doesn't want to be "the cheaper solution" as they put it, then there's some serious execution work that needs doing across all of their product stack.

 

[R0H1T]

You are deeply confused.

People may have referenced the original WCCF Tech article stating that, but find me someone who actually believed that.

Here is the thread, and not a single believer, yet you falsely claim people on this forum were peddling the new Conroe lie.

http://forums.anandtech.com/showthread.php?t=2425928

Doesn't the OP count ?

Secondly there is a good reason why that thread was based on a pile of BS, it's the same old uarch that's been going strong for a decade now. So even if theoretically possible it was next to impossible to get that level of performance, likewise as more leaks trickle we'll get to know how good or bad Zen is compared to Excavator.

That was the point I was making & btw you didn't reply to the rest of my post as to how Intel hyped up Broadwell, with many posters here chiming in, or do you deny that too?

This thread is basically all speculation at this point just like 90% of the threads on AT & yet supposedly the AMD supporters are the only ones that make tall claims right, no one from Intel or any supporting Intel on this forum has done that ever have they?

 

[AtenRa]

For cellphones and tablets, sure. Can you say the same thing about high power CPUs? I don't know where this confidence is coming from. How can you seriously use Intel as an example of clocks getting lower on finfet mobile oriented nodes, and then be convinced that AMD is immune to the same effect?

1. The Globalfoundries 14nm FF LPP can scale from low power to high performance devices. It is not be made for low power cellphones only.

2. ZEN is a new mArchitecture, It is designed for the 14nm FF from the start. Much like Broadwell and Skylake.

And if you have a look at Intels 22nm FF products, Haswell Refresh clocks (Core i7 4790K) are higher than 32nm SandyBridge (4GHz vs 3.5GHz) at lower TDP (with higher iGPU performance). And this was not an entire new mArch over SB like ZEN will be over BD.

Mid price mobile? 15W Broadwell i5s have caught up to 19W Kaveri in graphics (Tomb Raider, Bioshock, not 3DMark) and pull farther ahead in CPU performance. There's a gap between CT/Core that AMD can currently squeeze into, but if AMD doesn't want to be "the cheaper solution" as they put it, then there's some serious execution work that needs doing across all of their product stack.

15W TDP Core i5 Broadwell will go against Carrizo not Kaveri in 2015.

 

[CHADBOGA]

Doesn't the OP count ?

As he never stated he believed it, so clearly not.

Secondly there is a good reason why that thread was based on a pile of BS,

Yes, because it was a WCCFTech article.

That was the point I was making & btw you didn't reply to the rest of my post as to how Intel hyped up Broadwell, with many posters here chiming in, or do you deny that too?

What was Broadwell supposed to do that Haswell wasn't already doing?

And how about providing some actual quotes of ridiculous Broadwell claims by posters here.

You have already been proven wrong about posters here believing Skylake was another Conroe, you lack the credibility to be given the benefit of any doubt, so produce the quotes by forum members here, if they exist.

This thread is basically all speculation at this point just like 90% of the threads on AT & yet supposedly the AMD supporters are the only ones that make tall claims right, no one from Intel or any supporting Intel on this forum has done that ever have they?

The big difference is that AMD supporters are claiming AMD will achieve market share gains, 5 to 10 times higher than AMD themselves have stated is likely.

That is total detachment from reality.

It appears that despite being made fools of over Bulldozer, the same people are lining up again to predict super dooper things for AMD.

 

[mrmt]

Yeah there's definitely a market for high performance CPU in the enthusiast market. It fits perfectly with AMD's synergy given their high performance enthusiast class graphics capability. DX12, Mantle and Vulkan would have made that decision quite easy. A high performance, high core count/threaded ZenFX could very well be a gaming powerhouse.

Yeah, a *huge* market. That's why AMD is telling the guys that foot their bill tto expect a drop in revenue from this segment.

 

[ShintaiDK]

The big difference is that AMD supporters are claiming AMD will achieve market share gains, 5 to 10 times higher than AMD themselves have stated is likely.

Thats really one of the key points. I can understand that the dedicated supporters claim the very best and perhaps a little more than the company itself. That is usual often based on best case scenarios already.

But when the supporters starts to claim completely out of this world projections the company doesnt even remotely believe themselves. Then the chain have really fallen of. Then we are just over in fantasy land with little fairies and leprechauns.

 

[Abwx]

15W Broadwell i5s have caught up to 19W Kaveri in graphics (Tomb Raider, Bioshock, not 3DMark) and pull farther ahead in CPU performance.

Lol, quite a 15W Broadwell, 36.8W power comsumption for the laptop while the "19W" Kaveri Laptop is at 33.5W, obviously this "15W" BDW has higher TDP than a 19W Kaveri.

Tests are made in single channel, give thoses laptop a second RAM stick and you ll see wich is the GPU that will scale better with RAM bandwith increased, i guess that such half baked benches are necessary to keep Intel GPU allegedly competitive.

That said your post has some relevance because the AMD powered HP Elite Pro seems to have Carrizo instead of Kaveri in their next iteration, too bad this is a sign that it will be used in expensive laptops first, indeed according to a SA member who have genuine info there will be no 35W parts at launch, powers will be 12, 15 and 25W, this point to high efficency at preferably quite low powers.

 

[AtenRa]

Yeah, a *huge* market. That's why AMD is telling the guys that foot their bill tto expect a drop in revenue from this segment.

Revenue going down, profits going up. You know, they will stop selling all those low end CPUs and only concentrate on the market segments they will gain profits. High-end Desktop CPUs like ZEN 4-6-8 cores and APUs for Desktop and especially for Laptops.
The plan is to sell less volume but higher margin products that will give them profit. That is why they are saying PC Revenue will be flat for the coming quarters.

You did see the videos, im sure you heard that part did you ??

 

[mrmt]

So, while I don't see it exactly the same way as you, the dog & pony show on Analyst day does appear to be designed to cover up this reality.

They didn't really hide this from anyone. Actually Lisa wrote that the future is about focused ARM offers "ARM only where it makes sense", far different from the "ARM will win in the long run" statement that came under the duo Feldman/Read. She was also pretty blunt when talking about Skybridge, telling everyone that it didn't generate enough interest on the market, and she did not commit to Seattle ramp up despite the product being ready. This is as direct as she could be, especially because most of the team that was executing the old plan is still at AMD, and the BoD is the same.

I think the show was all about dismantling "ambidextrous" and going back to x86, but it also showed a weakness: K12 was supposed to be the lead product and since it was server optimized it also made Zen being server optimized. Basically AMD will again bet their farm on a server optimized architecture on the consumer market while lacking the funds optimize for it.

For AMD current strategy they would be far better suited to have a consumer optimized architecture and server products to get the low end of the server market, think a Xeon-D competitor + APU line leveraging on AMD graphics. But it seems that "ambidextrous" damaged this strategy and now AMD has to go after medium servers and *desktops*.

 

[mrmt]

Yes, the return to high performance FX processors means people won't have to settle for intel's meager couple percentage point performance increases. Competition will be good! (well, maybe not for AMD shorts, but no one cares about them anyway. )

I must say that Zen has all the elements of a dangerous product, perfectly suitable to short strategies for when AMD miss the quarterly sales forecasts. Wanna come aboard? :awe:

 

[mrmt]

There is no particular reason to think that AMD, with access to a better process in 2016 (14nm FinFET+), could not do 8 high-IPC cores at 4.0 GHz stock if they wanted to. AMD is traditionally more aggressive about running their chips (both CPUs and GPUs) closer to the margins than Intel.

There are. One is for the same reason Intel couldn't keep Conroe clocks at Prescott levels: A beefier core is harder to clock higher than an anemic core like Bulldozer, and the other is that the 14nm Samsung node isn't really optimized for this kind of clocks.

 

[R0H1T]

As he never stated he believed it, so clearly not.


Yes, because it was a WCCFTech article.


What was Broadwell supposed to do that Haswell wasn't already doing?

Oh so you want a signed statement is it, guess what no does that on AT & if the OP wasn't expecting such results (even for a sec) he would've stated his position in the first few pages itself wouldn't he? By the same token no one here has stated explicitly that Zen will be x times better than BD or y times more efficient, will sell z multiples more than what opterons are selling today.

If you want definitive statements maybe you should just ask the posters yourself if they stand behind their claim or not!

I expect 40% more IPC from Zen, as AMD stated, now you may take it as a prediction, expectation or guesstimate but I stand by what I said - that's as definitive as you get here.

The estimates I posted previously are my expectations & again I stand by them but I'm not the one behind Zen so I will not make a statement, since I have no knowledge of what the final product would be like.

And how about providing some actual quotes of ridiculous Broadwell claims by posters here.

You have already been proven wrong about posters here believing Skylake was another Conroe, you lack the credibility to be given the benefit of any doubt, so produce the quotes by forum members here, if they exist.

Oh hey look here now what would you call that - prediction, statement whatever else?

The big difference is that AMD supporters are claiming AMD will achieve market share gains, 5 to 10 times higher than AMD themselves have stated is likely.

That is total detachment from reality.

It appears that despite being made fools of over Bulldozer, the same people are lining up again to predict super dooper things for AMD.

No, AFAIK no one has put a number as to what AMD will or will not achieve in the next two years, as for me I don't even know what AMD said their expectations was, so for you to say 5~10 times (boy that's some margin you've got covered) marketshare gains were predicted is disingenuous to say the least.

 

[ShintaiDK]

Oh hey look here now what would you call that, prediction, statement whatever else?

Did you even read the post you linked?

No, AFAIK no one has put a number as to what AMD will or will not achieve in the next two years, as for me I don't even know what AMD said their expectations was, so for you to say 5~10 times (boy that's some margin you've got covered) marketshare gains were predicted is disingenuous to say the least.

Actually they did. People even started to dream about 30-40% PC marketshare if you look back in the thread. AMD expects flat to negative.

 

[R0H1T]

Did you even read the post you linked?

Yes, a 65W BDW chip, at stock I presume, beating a ~90W Haswell part clocked to its limit? The answer is obvious isn't it, unless the poster was hoping for some magic results

You can find similar statements in this thread & you'll vilify them plus hang the poster, so why these double standards now?

 

[ShintaiDK]

Yes, a 65W BDW chip, at stock I presume, beating a ~90W Haswell part clocked to its limit? The answer is obvious isn't it, unless the poster was hoping for some magic results

You can find similar statements in this thread & vilify them plus hang the poster so why double standards now?

Nobody claimed such thing in the post you linked.

 

[R0H1T]

Nobody claimed such thing in the post you linked.

So I guess the poster was just trolling?

 

[AtenRa]

Did you even read the post you linked?



Actually they did. People even started to dream about 30-40% PC marketshare if you look back in the thread. AMD expects flat to negative.

Where did you see AMD talking about marketshare ??

 

[cytg111]

1. The Globalfoundries 14nm FF LPP can scale from low power to high performance devices. It is not be made for low power cellphones only.

2. ZEN is a new mArchitecture, It is designed for the 14nm FF from the start. Much like Broadwell and Skylake.

And if you have a look at Intels 22nm FF products, Haswell Refresh clocks (Core i7 4790K) are higher than 32nm SandyBridge (4GHz vs 3.5GHz) at lower TDP (with higher iGPU performance). And this was not an entire new mArch over SB like ZEN will be over BD.

You state it as "GF 14nm FF LPP" is a proven fact in history. Lots of things can go wrong from here till then on the process side of things and add to that GF's track record. Samsung, i know i know.. Just saying a hell of a lot of pieces has to line up to pull this off.. Not impossible but far far from trivial either.

Where did you see AMD talking about marketshare ??

- Ignore him and mrmt, they are not here to debate you.

 

[ShintaiDK]

So I guess the poster was just trolling?

I think its you being the one with the problem.

None of the 2 you link to in the post claims, predict or state anything ridiculous.

 

[ShintaiDK]

Where did you see AMD talking about marketshare ??

Yes its not directly marketshare. But its revenue and margins. And obviously AMD is not going to sell cheap chips for pennies to gain any share due to margins.