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

[MiddleOfTheRoad]

Thats your vision of the future AMD?

Why not? Matrox is consistently profitable. A leaner and more agile AMD. Less product offerings, but better performance for each of them. I believe this is where they are heading already.

 

[ShintaiDK]

Why not? Matrox is consistently profitable. A leaner and more agile AMD. Less product offerings, but better performance for each of them. I believe this is where they are heading already.

So a goodbye to the PC segment, servers and so on? And all focus to niche and semicustom?

Without a leading edge R&D funding, you are going to be severely limited in the market options.

 

[JDG1980]

AMD is already getting hammered for selling their 140W-220W chips.

They're getting hammered because these 140W-220W chips can't compete with Intel's 77W-95W offerings (including Sandy Bridge chips dating back to 2011). If Zen + GCN 2.0 + 16nm FinFET enables AMD to get the equivalent of an i5-2600K plus a good midrange GPU (something like fully enabled Tonga) plus 8GB of shared HBM, all in a 150W power envelope, that's a far different product and is going to be a lot more successful. As others have pointed out, once you stop having to cool about 5 different components and put all the heat-generating stuff on one chip, that opens the door to all kinds of new form factors. The Mac Pro and the Streacom fanless heatpipe cases offer examples of how this could be done.

 

[monstercameron]

Because they are part of a specialized subset of the market that requires high performance?



Some people don't care about heating or size or anything like that. However, the vast majority of the consumer market (those not buying a mac pro for video editing or photoshop or the like) do not want huge power guzzlers.



There is still the top of the pyramid...but the base is so much larger.

Even assuming you are correct on what consumers don't want it should be noted that tdp isn't constant. Another note is that consumers don't really know what they want Until they need it.

Attempting a car analogy:
It's like a sports car that can start stop its engines while doing nothing and enabling/disabling cylinders depending on how fast they wanna go.

 

[VirtualLarry]

Or will it be the gamer that will accept a mediocore at best CPU paired with whatever left of the GPU that may be in 2016+?

Pessimism? Or trolling?

Its at best a niche product without any chance in the consumer segment. APUs today is a sales flop. Adding a bigger GPU and increasing TDP to crazy levels isnt going to change that. Its only going to make it worse.

Not at all. It offers AMD the opportunity to scale up, go big, make APUs make a real difference in terms of performance. Right now, they are in a sort of no-man's land, performance-wise. Better IGP than Intel, but not good enough for gaming, really, unless you like 800x600 low.

I would much prefer an APU setup on a plugin card at 300W, if the performance is there.

 

[monstercameron]

Pessimism? Or trolling?





Not at all. It offers AMD the opportunity to scale up, go big, make APUs make a real difference in terms of performance. Right now, they are in a sort of no-man's land, performance-wise. Better IGP than Intel, but not good enough for gaming, really, unless you like 800x600 low.



I would much prefer an APU setup on a plugin card at 300W, if the performance is there.

I agree with your sentiment but I have to point out the misleading part about 800x600 low. As I have pointed out numerous times in the past, Kaveri is suitable for the majority of gamers out there. Save for AAA games and game worlks titles, Kaveri will most likely give you a good experience.

 

[cytg111]

Pessimism? Or trolling?

- Really? You have to ask?

 

[ShintaiDK]

Pessimism? Or trolling?

Try realism.

I would much prefer an APU setup on a plugin card at 300W, if the performance is there.

You have to ask, how many would buy it besides potentially you. If it cant sell then it doesnt matter.

 

[MiddleOfTheRoad]

I agree with your sentiment but I have to point out the misleading part about 800x600 low. As I have pointed out numerous times in the past, Kaveri is suitable for the majority of gamers out there. Save for AAA games and game worlks titles, Kaveri will most likely give you a good experience.

I agree with MonsterCameron. An A10 Kaveri is already good enough for some gamers. Granted, the Kaveri can't quite manage high detail settings at 1080p for the latest games.

The gamers that are satisfied playing in medium detail settings..... Kaveri can do a pretty good job for those buyers.

 

[Enigmoid]

Even assuming you are correct on what consumers don't want it should be noted that tdp isn't constant. Another note is that consumers don't really know what they want Until they need it.

Attempting a car analogy:
It's like a sports car that can start stop its engines while doing nothing and enabling/disabling cylinders depending on how fast they wanna go.

But the form factor needs to be designed for that peak power. Believe me, consumers hate stuff overheating and shutting off.

Now of course it is possible to cool these kinds of components but the cost is higher.

But lets not forget another barrier: OEMs

Before the average consumer even sees a 300W APU the OEM has to decide to build a product that will use it. Will OEM's embrace 300W APUs? Perhaps, perhaps not. But they will be a new thing on the market and this will require some marketing $$$ to get the information to the consumer.

IMO, while an APU may be a good idea for a laptop, I can't see myself buying one for a desktop. Buy a modern CPU today and it will pretty much last you 5+ years (SB should easily be good for the next 3 years, especially if overclocked). Meanwhile, pretty much any GPU from that era will be hopelessly obselete. So while the CPU may last a while, the igp will be pretty much useless after 3 years.

 

[cbn]

I agree with MonsterCameron. An A10 Kaveri is already good enough for some gamers. Granted, the Kaveri can't quite manage high detail settings at 1080p for the latest games.

The gamers that are satisfied playing in medium detail settings..... Kaveri can do a pretty good job for those buyers.

The problem with those APUs is the price compared to something like the Athlon x 4 860K and a Power color R7 250X (which often goes on sale for $59.99 AR FS. Normal price appears to be $79.99 AR FS).

So one question is how will this new high TDP APU compare to the current scenario when have? Will it be a better deal or worse value compared to dCPU and dGPU? (Remember when comparing APUs to dCPU and dGPU we have to factor in CPU throttling as a reducer of value for the APU as well.)

 

[monstercameron]

But the form factor needs to be designed for that peak power. Believe me, consumers hate stuff overheating and shutting off.



Now of course it is possible to cool these kinds of components but the cost is higher.



But lets not forget another barrier: OEMs



Before the average consumer even sees a 300W APU the OEM has to decide to build a product that will use it. Will OEM's embrace 300W APUs? Perhaps, perhaps not. But they will be a new thing on the market and this will require some marketing $$$ to get the information to the consumer.



IMO, while an APU may be a good idea for a laptop, I can't see myself buying one for a desktop. Buy a modern CPU today and it will pretty much last you 5+ years (SB should easily be good for the next 3 years, especially if overclocked). Meanwhile, pretty much any GPU from that era will be hopelessly obselete. So while the CPU may last a while, the igp will be pretty much useless after 3 years.

You bring up very good points and apus do have a bit of obsolescence by design but I will counter by stating that you can still augment your apus graphics performance with a dgpu. On a more specific note Kaveri might last longer than expected due to the new apis, it is still on a watch and see basis.

 

[frozentundra123456]

The problem with those APUs is the price compared to something like the Athlon x 4 860K and a Power color R7 250X (which often goes on sale for $59.99 AR FS. Normal price appears to be $79.99 AR FS).

So one question is how will this new high TDP APU compare to the current scenario when have? Will it be a better deal or worse value compared to dCPU and dGPU? (Remember when comparing APUs to dCPU and dGPU we have to factor in CPU throttling as a reducer of value for the APU as well.)

Yea, the problem is not whether Kaveri is good enough (in my opinion it isn't quite there) but even if for the sake of argument you say it is, the question is whether there is a similarly priced better alternative, and for the desktop the answer is clearly yes. Will HBM change that? Perhaps, but that is yet to be seen.

 

[JDG1980]

IMO, while an APU may be a good idea for a laptop, I can't see myself buying one for a desktop. Buy a modern CPU today and it will pretty much last you 5+ years (SB should easily be good for the next 3 years, especially if overclocked). Meanwhile, pretty much any GPU from that era will be hopelessly obselete. So while the CPU may last a while, the igp will be pretty much useless after 3 years.

I'm not sure this is the case. TSMC's stagnation on 28nm has made GPU lifecycles much longer. If you bought a Radeon HD 7970 back in December 2011, that card would still be a perfectly viable mid-range GPU today. (After all, AMD sells the same card today - with slightly higher clocks and worse power consumption - as the R9 280X.) And the odds are that we're going to be stuck on 16nm FinFET for at least as long as 28nm, if not longer.

Yea, the problem is not whether Kaveri is good enough (in my opinion it isn't quite there) but even if for the sake of argument you say it is, the question is whether there is a similarly priced better alternative, and for the desktop the answer is clearly yes. Will HBM change that? Perhaps, but that is yet to be seen.

Lack of memory bandwidth is the reason why APUs currently are not viable. Kaveri reviews indicated that there wasn't much improvement on most games when going from 384 SPs to 512 SPs; the bottleneck was memory bandwidth. DDR3 is just not going to cut it. The original intent was for Kaveri to support GDDR5 (probably incorporated onto the motherboard), but that was cut. The advent of HBM changes the whole equation; once it's possible to put 8GB of ultra-high-bandwidth memory on top of the APU, then devoting a larger portion of the die to more shaders starts to scale again, just as it does with discrete cards.

Currently, none of AMD's APUs can even compare with the old Radeon HD 7750, a low-end card released in February 2012 (over three years ago) that cost $110 at that time. That's a pretty pathetic level of performance. Since 1080p is still the most popular resolution, it will make a big difference when APUs get strong enough to play AAA titles at Full HD on normal detail settings. Right now, that means something similar to GM204 or Tonga, maybe a little less. 16nm FinFET combined with HBM should make that possible.

 

[frozentundra123456]

True, but the cpu performance is nothing to write home about either. So if you remove the bandwidth restriction, the cpu might become the limiting factor. Not to mention that by the time we see 14nm APUs with HBM, we should have somewhat faster cpus and a new generation of much faster dgpus.

 

[tynopik]

Not to mention that by the time we see 14nm APUs with HBM, we should have somewhat faster cpus

yeah, another 5% gain, woohoo!

Intel's performance stagnation will allow AMD to catch up to relevancy

 

[Gikaseixas]

If AMD releases an APU with 2500K and Radeon 270X performance in one package that consumes 300 watts i think many buyers accept such products.

 

[SimianR]

I think price is a big factor. If this shaved a decent chunk off the cost of building a new system then it might actually be a worthwhile option.

 

[DrMrLordX]

It is, although it's easier to run straight x86 code than to rewrite everything for OpenCL. Of course Knights Landing should also be able to do 64-bit FP at 1/2 the 32-bit rate vs. the 290x doing 64-bit at 1/8 the 32-bit rate.

Right, but remember, a 290x is just a consumer card, so 64-bit is deliberately gimped. But that was already covered elsewhere in this thread so . . .

Anyway, the beauty of Knight's Landing is in the fact that you don't have to go to OpenCL at all. You can boot off the Knight's Landing chip and run code that's been autovectorized for you by ICC and Bob's your uncle, it works. A <insertvideocardhere> requires a host CPU plus something like OpenCL or CUDA.

AMD is already getting hammered for selling their 140W-220W chips. You still think a 300W chip today would be welcomed by the average consumer and OEM?

OEM, maybe not. It would depend on the market segment. Performance desktop? Sure, why not? Not many of those sell anymore, but you can still get "gamer" desktops from OEMs and custom builders if you buy from the right places.

If AMD releases an APU with 2500K and Radeon 270X performance in one package that consumes 300 watts i think many buyers accept such products.

Bing, we have winnar. I am thinking the GPU performance could be above that of a 270X though . . . we're talking a max ~200W TDP assigned to GCN cores implemented on Samsung/GF 14nm. That's a lot of GCN cores.

 

[TurtleCrusher]

Bing, we have winnar. I am thinking the GPU performance could be above that of a 270X though . . . we're talking a max ~200W TDP assigned to GCN cores implemented on Samsung/GF 14nm. That's a lot of GCN cores.

14/16nm? 200w? We're likely talking 1.25 to 1.5x of a 290x.

 

[Insert_Nickname]

AMD is already getting hammered for selling their 140W-220W chips. You still think a 300W chip today would be welcomed by the average consumer and OEM?

They're getting hammered because these 140W-220W chips can't compete with Intel's 77W-95W offerings (including Sandy Bridge chips dating back to 2011). If Zen + GCN 2.0 + 16nm FinFET enables AMD to get the equivalent of an i5-2600K plus a good midrange GPU (something like fully enabled Tonga) plus 8GB of shared HBM, all in a 150W power envelope, that's a far different product and is going to be a lot more successful. As others have pointed out, once you stop having to cool about 5 different components and put all the heat-generating stuff on one chip, that opens the door to all kinds of new form factors. The Mac Pro and the Streacom fanless heatpipe cases offer examples of how this could be done.

I couldn't have put it better myself.

Besides I don't see anyone complaining of 200W+ TDP GPUs. Or PS4/Xbone APUs for that matter. So long as the performance is there to justify TDP people won't care. As for niche product, discrete GPUs are that already.

I could also see a potential market for such a 200W+ APU in the HPC space, if the performance is good enough.

 

[DrMrLordX]

14/16nm? 200w? We're likely talking 1.25 to 1.5x of a 290x.

I just didn't wanna overstate my case. AMD has dropped the "200-300W" number so, assuming ~100W for the CPU (which would be a lot of Zen cores on its own), we're looking at 100-200W for the iGPU.

But yeah, 270x is old hat by 2016 standards, using a process that is/will be similarly dated. It really would not be that hard for AMD to eclipse that performance level with such a "healthy" power envelope, even if it only leaves 100W for the iGPU.

 

[Fjodor2001]

Just a thought: If AMD pulls off Zen being a significant x86 improvement at Sandy Bridge or above performance, at 14 nm, and with chips using HBM and latest GPU cores, then I think it's quite obvious it could be a massive success. Note that I'm not only talking about this 200-300 W SKU, but also about lower TDP SKUs that likely also will be on AMD's repertoire.

Having this in mind, isn't it strange that this isn't reflected in AMD's stock price?

I see two options:

1. AMD will deliver something at least close to what I mentioned above and has been discussed in this thread => AMD's stock is seriously undervalued.

2. AMD will deliver something completely different from 2) (e.g. CPU performance close to Excavator or similar) => What has been claimed by AMD is completely false.

 

[jpiniero]

Just a thought: If AMD pulls off Zen being a significant x86 improvement at Sandy Bridge or above performance, at 14 nm, and with chips using HBM and latest GPU cores, then I think it's quite obvious it could be a massive success. Note that I'm not only talking about this 200-300 W SKU, but also about lower TDP SKUs that likely also will be on AMD's repertoire.

Who exactly are you expecting to buy this? I mean they could theoretically net some HPC deals out of it but that might be it.

 

[Fjodor2001]

Note that I'm not only talking about this 200-300 W SKU, but also about lower TDP SKUs that likely also will be on AMD's repertoire.

Who exactly are you expecting to buy this? I mean they could theoretically net some HPC deals out of it but that might be it.

The 200-300 W SKUs: Anyone shopping for mig/high end a gaming computer.

Lower TDP SKUs (45-95 W): Anyone shopping for a desktop APU (currently considering a Core i3-i7 or similar). If AMD's APUs are close to SB performance and at lower price, it'll sell. SB level of performance is good enough for most people. And likely they'll get a better iGPU with AMD.