AMD 12 Core A-Series APU Possibly Inbound

I doubt it means anything. They are 12 cores, AKA "compute cores". 4 CPU cores + 8 GPU cores. Remember that top-end Kaveri was advertised as a "12 compute core" product.

Also multi-processor APU's currently aren't possible. There's no fast interconnect tech like HyperTransport available on those platforms to allow for such a thing.
 
Last edited:
True, but reading the article, I agree with their logic that perhaps this indicates steamroller cores and not just generic "compute" cores.
 
I doubt it means anything. They are 12 cores, AKA "compute cores". 4 CPU cores + 8 GPU cores. Remember that top-end Kaveri was advertised as a "12 compute core" product.

Agreed. AMD is just looking to alter perceptions with marketing

Not that it really matters. ¯\_(ツ)_/¯
 
APUs are a great idea, and having both CPU and GPU working together to solve problems greatly increases the value of the solution. i have seem cluster of APUs winning college competitions of cluster building inside a budget.

Intel has tried hard to leapfrog AMD on integrated GPU performance, but funny enough it is compute and better drivers that in the end made APUs such good chips.
 
Well I don't mind an APU as long as it's more powerful than my current FX-9370 at CPU chores.
 
True, but reading the article, I agree with their logic that perhaps this indicates steamroller cores and not just generic "compute" cores.

One thing that they didn't consider is how horrible the power draw and clock speed hits would be. Current Kaveri is ~95W TDP with just four cores clocked at 4.0 GHz turbo. 12 cores on a single die on the same process would be catastrophic. Maybe the clocks could be dropped to 1.8 GHz or something, but that'd be about it, lol.

A big-die product (well, bigger than the 245mm^2 that Kaveri already is, anyway) with SR cores would theoretically only come from an Opteron successor to Warsaw on G34. Socket FM2+ doesn't seem to support any crazy configs like octocores or anything, let alone 12-cores, at least not at the clockspeeds that enthusiasts would expect.

Now there is a rumor floating around that Carrizo may not hit the desktop at all: www.bitsandchips.it/hardware/9-hardware/4643-socket-fm2-fino-al-2016-niente-ddr4-per-le-apu-desktop
 
twelve cores sure.

but 4 CPU + 8 GPU...

why not 8 CPU + 4 GPU?

47% of the kaveri die space is GPU:
http://www.extremetech.com/wp-conten...d-to-intel.jpg
if you consider that roughly 20% is uncore, that leaves roughly 33% as CPU.

give or take, 8 shader cores is fifty percent larger than 4 cpu cores.
you could double that cpu portion to 66%, and still leave 14% for shader cores.

make the total die size just 10% bigger and you have an 8 cpu core APU with 4 HSA enabled shader cores ready to grind through FPU work. pretty much die-size neutral.

wouldn't that be worthy of an "FX" title?
 
It would be worthy of FX, but AMD doesn't really care about that (and neither do their target audiences for these chips, that being the "average" people.) HSA and OpenCL benefits mostly from potent GPU strength. I guess AMD figured that 4 CPU cores was the perfect median for a balance between CPU grunt and GPU power.

Keep in mind that AMD's desktop/mobile methodology is pretty much the same as Intel's: Just like how all of the i3/i5/i7's on the consumer platform are identical silicon to their mobile counterparts, the A-series chips are the same ones that get put in the mobile notebooks (besides the ultra-low power stuff such as Beema/Mullins and such.) So no one who uses a laptop will ever realistically need 6 or 8 CPU cores, which is the same reason why all those naive idealists who constantly clamor for Intel to release hexacore parts on the consumer platform will never have their wet dreams satisfied.

It is not a far cry to say that all of the A-series stuff is primarily designed for mobile first, then the rest of the chips that didn't make it into mobile products get put on the desktop, same with Intel and their i-whatever series on their consumer desktop platform.

So while in theory a lot of these CPU/GPU configs make perfect sense, in practice they won't happen for now.
 
One thing that they didn't consider is how horrible the power draw and clock speed hits would be. Current Kaveri is ~95W TDP with just four cores clocked at 4.0 GHz turbo. 12 cores on a single die on the same process would be catastrophic. Maybe the clocks could be dropped to 1.8 GHz or something, but that'd be about it, lol.

A big-die product (well, bigger than the 245mm^2 that Kaveri already is, anyway) with SR cores would theoretically only come from an Opteron successor to Warsaw on G34. Socket FM2+ doesn't seem to support any crazy configs like octocores or anything, let alone 12-cores, at least not at the clockspeeds that enthusiasts would expect.

Now there is a rumor floating around that Carrizo may not hit the desktop at all: www.bitsandchips.it/hardware/9-hardware/4643-socket-fm2-fino-al-2016-niente-ddr4-per-le-apu-desktop

That's an interesting assertion, if it is indeed a twelve steamroller core with GCN as well, then power and heat are huge issues. But perhaps it's an pay with the graphics disabled? 2 GCN compute units not counted used basic duties if no dedicated card is there? It's still a huge die increase regardless.
 
Keep in mind that AMD's desktop/mobile methodology is pretty much the same as Intel's: Just like how all of the i3/i5/i7's on the consumer platform are identical silicon to their mobile counterparts, the A-series chips are the same ones that get put in the mobile notebooks (besides the ultra-low power stuff such as Beema/Mullins and such.)

Except Intel aren't doing that. Which is to say that intel now have a significant delay between the arrival on 14nm broadwell on mobile and desktop.

And last week we saw an AMD slide on Carizzo, where I said something along the lines of; "hey look chaps, isn't it odd that they have only put a PCIe3.0 8x graphics bus on their shiny new Excavator APU! What could that possibly mean?"

and now it appears there are rumours that Carizzo is mobile only (quelle surprise):

http://www.fudzilla.com/home/item/35337-rumour-amd-carizzo-will-be-mobile-only

which would rather indicate they must have different plans for the desktop, plans involving more PCIe lanes (remember Kaveri has 24x), and presumably higher performance parity with Intel in order to keep the average selling price reasonable.

i ask you again, given this, what is unreasonable about the prospect of a 12 core (reversed) kaveri respin?
 
You'd be hard-pressed to actually fully saturate the bandwidth of even 8x PCIe 3.0. The delay between Broadwell mobile and desktop doesn't change anything. Both Llano and Trinity launched in mobile months before they came to the desktop. Kaveri was an anomaly in that it hit desktop first and mobile came last.

which would rather indicate they must have different plans for the desktop, plans involving more PCIe lanes (remember Kaveri has 24x), and presumably higher performance parity with Intel in order to keep the average selling price reasonable.

Carrizo will hit desktop and be marginally better than 95W Kaveri. It won't be majestical or anything.

i ask you again, given this, what is unreasonable about the prospect of a 12 core (reversed) kaveri respin?

Refer to this post I made earlier, for starters. Then think about how AMD doesn't have the money to squander on having another production line for a series of CPU's that would only appeal to a niche market and not sell as well as their APUs do. Even this 12-core APU you propose would be a terrible product, because they wouldn't go back to 32nm SOI for it, and it would have much lowered clocks that would invalidate any IPC gain it'd have. The chip would be pointless.
 
You'd be hard-pressed to actually fully saturate the bandwidth of even 8x PCIe 3.0. The delay between Broadwell mobile and desktop doesn't change anything. Both Llano and Trinity launched in mobile months before they came to the desktop. Kaveri was an anomaly in that it hit desktop first and mobile came last.

Carrizo will hit desktop and be marginally better than 95W Kaveri. It won't be majestical or anything.

Refer to this post I made earlier, for starters. Then think about how AMD doesn't have the money to squander on having another production line for a series of CPU's that would only appeal to a niche market and not sell as well as their APUs do. Even this 12-core APU you propose would be a terrible product, because they wouldn't go back to 32nm SOI for it, and it would have much lowered clocks that would invalidate any IPC gain it'd have. The chip would be pointless.

you cannot sell 8x PCIe 3.0 lanes for anything more than <£60 market.
not when Intel will sell you an Core based product with 16x
and not when your previous product had 24x.

the rumour is that Carizzo won't hit the desktop, at least not in its current incarnation.

the point is that amd IS squandering money right now, its ASP is being crushed while its still making the same big 32nm chips that must be have frightening little margin (an 8350 was going for £110 the other day on amazon).
they need higher-end products that compete higher up the price scale.
You said:
"One thing that they didn't consider is how horrible the power draw and clock speed hits would be. Current Kaveri is ~95W TDP with just four cores clocked at 4.0 GHz turbo. 12 cores on a single die on the same process would be catastrophic. Maybe the clocks could be dropped to 1.8 GHz or something, but that'd be about it, lol."
Well yes. it had ~1.5b extremely process dense shader transistors, that were highly clocked to differentiate it from Intel iGPU's. No surprise.
 
Last edited:
you cannot sell 8x PCIe 3.0 lanes for anything more than <£60 market.
not when Intel will sell you an Core based product with 16x
and not when your previous product had 24x.

There's no guarantee that desktop Carrizo will be limited to 8x PCIe 3.0. Not to mention that anything above Core i3 is irrelevant since that's what these APU's are mainly competing against.

the rumour is that Carizzo won't hit the desktop, at least not in its current incarnation.

I looked into the rumor and it's pretty stupid. Carrizo can only miss FM2+ if they decide - for some unknown reason -- to cancel Toronto as well. Since there's no indication of them doing that, it's safe to say that Carrizo will hit desktop as has been planned for quite some time.

they need higher-end products that compete higher up the price scale.

Indeed they do. But they can't just do some black magic and come up with higher-end products that sane people are willing to buy.

Well yes. it had ~1.5b extremely process dense shader transistors, that were highly clocked to differentiate it from Intel iGPU's. No surprise.

Onboard graphics or not, 28nm SHP was tailored for a balance between CPU/GPU. It's not good for clocking CPU's high like 32nm SOI was.
 
So no one who uses a laptop will ever realistically need 6 or 8 CPU cores, which is the same reason why all those naive idealists who constantly clamor for Intel to release hexacore parts on the consumer platform will never have their wet dreams satisfied.

How did you become so self-satisfied?

"Intel holding out on upping core count due to zero competition from AMD? I know! I'll make fun of people who refuse to keep quiet about it!"

The simple fact of that matter is that Intel could make tidy profits selling unlocked 8 core parts at 4+ GHz for less than $300. They simply haven't pushed the R&D necessary to do it, at all.
 
Did you get personally offended by that remark? It should be obvious that the mobile market is more important than the desktop DIY channel is. Why would Intel release unlocked octocores for a (relatively) low price when all that would do is cannibalize the shit out of the rest of their processor lineup? May as well make people who want that sort of thing shell out over a grand to get it.

It's like I said in my original post. Those consumer i3/i5/i7 chips are the same things that get put in most of the mobile SKU's that use their "big cores". When you see hexacores in laptops from Intel, you'll see them on the desktop. Seeing as how that's incredibly naive and unrealistic (because no average joe needs that at all), it's safe to say it won't happen. Not even Intel with their insanely high R&D budget have ever made parts like that just for HEDT customers.
 
I've learned to never take anything that website says seriously. They're almost always way fucking off.

I'll believe it when I see it, but adding lots more cores is all fun and dandy, but you also need some more IPC improvements.
 
The norm on Intel seems to be dual core. (consumer-side).

It'll be a while before that shifts entirely to quad core, and even longer to shift all laptops to Intel-quad. So yeah, GFL getting a mainstream 6-8 core intel chip.
 
If you know that 8 jaguar cores and a decent gpu on one chip amounts to 125 Watt already. Guess what suddenly drops out of the realm of possibility ?

Yes 12 Steamroller cores would have a hard time running anything but AM3+ and still that means that there needs to be some decent mainboard and very good power support for this to operate. FM2+ is still limited to 100 Watt.

So what is not happening a 12 core APU :) on the FM2+ platform. You might have noticed one of the last "scares" that Roy Taylor (?) made on twitter when there was mentioned something about a new FX CPU ?

AMD just did not want to put any effort into anything for these last few generations so if you see stuff and it is not on the roadmap then it prolly is not something new and exciting.
 
Last edited:
How did you become so self-satisfied?

"Intel holding out on upping core count due to zero competition from AMD? I know! I'll make fun of people who refuse to keep quiet about it!"

The simple fact of that matter is that Intel could make tidy profits selling unlocked 8 core parts at 4+ GHz for less than $300. They simply haven't pushed the R&D necessary to do it, at all.

I'm not to sure what this is about ? 8 cores are you confused with 4C/8T ? That is what Intel has beside the 6C/12T which is rather expensive ?

Is Intel not the company that decided to cheap out on the materials used for conducting the main lid on the cpu ?
 
If you know that 8 jaguar cores and a decent gpu on one chip amounts to 125 Watt already. Guess what suddenly drops out of the realm of possibility ?

Yes 12 Steamroller cores would have a hard time running anything but AM3+ and still that means that there needs to be some decent mainboard and very good power support for this to operate. FM2+ is still limited to 100 Watt.

So what is not happening a 12 core APU :) on the FM2+ platform. You might have noticed one of the last "scares" that Roy Taylor (?) made on twitter when there was mentioned something about a new FX CPU ?

hmmm, might that have a hell of a lot to do with a high-speed 256bit DDR5 interface and 1152 shaders clocked at 800MHz? we now have mullins chips with four jaguar cores that run at console speed with a budget in mere watts.

WFFCtech may be daft enough to not understand how AMD markets its APU products, but why should we live within their limitations?

we already have a 12 'core' APU on FM2+, all we're talking about here is an internal rearrangement of what they provide: 4c+512s / 6c+384s / 8c+256s - all are 12 cores.

----------------------------------------------------

none of this is advanced tech, or using niche and bespoke solutions for marginal purposes.
it is bog standard steamroller cores, as used in kaveri
it is bog standard GCN shaders, as used in kaveri
it is bog standard 128b DDR3 2133 interface, as used in kaveri
it is bog standard uncore elements, as used in kaveri

less of one, more of the other. more or less die-size neutral, useing less transistor dense shaders units that eat thermal budget. allowing a higher ASP and the ability to cross-sell their APU's and attendant platforms with their GPU's.

A 100W power budget is plenty to provide a 12 'core' APU using a mature 28nm process.
 
Well the FX-8350 is a 8 core Steamroller cpu.rated 140 Watt. And for the platform FM2+ it does not make any sense at all.

So how do you want to do this? Run it at 3ghz and hope it does not go beyond the power envelope of FM2+?

Adding 2 more modules to a FX 8350 would somehow work below 100Watt?
Where Jaguar would be able to come close to lower power usage means that this just would not fly unless it is crippled beyond what is normally expected.
 
wrong.

the 8350 is a 125W processor, despite being built on the aging 32nm process.

what does not make sense about the FM2+ platform?
every kaveri to date conforms to the 95W envelope, despite having ~1.5bn transistor dense (and thereforwe thermally dense), GPU chewing through watts.

they have also quite happily created an A10 product at 65W that has a base speed of 3.5GHz... despite having 512 fully enabled shaders.

of course they can make a 100W product (the platform supports old 32nm FM2 parts), that uses more cores and fewer shaders, while still keeping the clock-speed around 4GHz.

with the steamroller uarch that will surpass the performance of the 8350 which will, along with HSA and a modern platform, easily justify an ASP in the i5 % range (~£180).

would you like your 12 'core' APU as a 4+8 or an 8+4 product, Sir?

i'm not saying this is a certainty, i am simply rebutting the daft arguments that:
1. it a thermal impossibility
2. that it is too bespoke to justify the development cost
3. that it is unnecessary from a sales strategy

how do you not understand this?
 
Even if they made other SKUs with those configs, hardly anyone would buy them. They wouldn't be put into laptops, and they'd sell only to AMD diehards in the DIY channel. As it stands they have one piece of "master silicon" that stretches all the way from server down to mainstream desktop and consumer mobile. The only differences were things being disabled (CU's, caches, etc.) Why would they do a separate production line for such a niche product that would barely sell? Hint: it's probably the reason they haven't done it and why they won't do it.

And I initially dismissed the rumor that Carrizo won't hit desktop, but there are more sources popping up in the whisper channel saying the same thing, so FM2+ might be another platform that got abandoned early by AMD. If true, AMD will have two desktop platforms that hardly anyone has any good reason to go to, forcing everyone to go Intel no matter what. GG no re AMD management.
 
how do you not understand this?

AMD clearly put all the eggs in Jim Keller's basket. They will not produce any effort to make a new product for the FM2+ which is not in the roadmap.

Your config of 4+8 and 8+4 sounds "okay". But the drawback is marketing and explaining where it needs to go from there.

Don't get me wrong 8 steamroller cores and and 4 graphics cores sounds like a very good deal but sadly AMD will not do things this way.
 
if this is true then AMD has real problems, for it's business sectors are no longer complementary and self-supporting to each other.
 
Legit Reviews reviewed two of the new APUs. They have switchable TDP (45w/65w), which is interesting. For the most part, AMD's APUs are competitive with the Pentium G3258 and punch above their weight when it comes to graphics tests.
 
It probably means there's an new APU called "A12". CPU marketing hasn't been based in reality since they switched from clock speeds to model numbers.

If they add CPU or GPU cores, it's going to be for the entire line of APUs (or desktop / server CPUs) and we'd have seen that coming for awhile. Investors and OEMs need to know that shit.

PS. that website is complete crap. You don't need to wait for them to be proven wrong, you can just look at their reasoning and baseless speculation.
 
Last edited:
Nice to know there's future upgrade path. Very happy with the A10-7850K. Very solid as a server pulling only about 40W and very capable playing BF4 at high 40 fps without dGPU.
 
Nice to know there's future upgrade path. Very happy with the A10-7850K. Very solid as a server pulling only about 40W and very capable playing BF4 at high 40 fps without dGPU.

Can't do it myself man, ECC.
I have no idea why they included certain server features like IOMMU in FM2(+) but not ECC.

Edit: the GPU cores I guess...
 
Last edited:
It would be worthy of FX, but AMD doesn't really care about that (and neither do their target audiences for these chips, that being the "average" people.) HSA and OpenCL benefits mostly from potent GPU strength. I guess AMD figured that 4 CPU cores was the perfect median for a balance between CPU grunt and GPU power.

Keep in mind that AMD's desktop/mobile methodology is pretty much the same as Intel's: Just like how all of the i3/i5/i7's on the consumer platform are identical silicon to their mobile counterparts, the A-series chips are the same ones that get put in the mobile notebooks (besides the ultra-low power stuff such as Beema/Mullins and such.) So no one who uses a laptop will ever realistically need 6 or 8 CPU cores, which is the same reason why all those naive idealists who constantly clamor for Intel to release hexacore parts on the consumer platform will never have their wet dreams satisfied.

It is not a far cry to say that all of the A-series stuff is primarily designed for mobile first, then the rest of the chips that didn't make it into mobile products get put on the desktop, same with Intel and their i-whatever series on their consumer desktop platform.

So while in theory a lot of these CPU/GPU configs make perfect sense, in practice they won't happen for now.

The interesting aspect of these chips comes when more games with mantle and dx12 come online. We've already seen examples of 8350s performing identically to i7-4770s.

Imagine amd releasing an apu with 6 modules / 12 cores for around 300 dollars that performs identically to intels 600 dollar 6 core i7 models. With better integrated graphics built in that can actually meaningfully assist the discreet gpu if it's all AMD.

I feel like I'm taking crazy pills, but are we headed into a world where intels process lead and faster ipc don't actually make a difference in future games? If we are, that could put amd back in the running on the desktop for gamers.
 
If cores don't show in task manager or some place else, i don't count them, also virtual cores does not count too. Half cores? I don't count them either.
 
I feel like I'm taking crazy pills, but are we headed into a world where intels process lead and faster ipc don't actually make a difference in future games? If we are, that could put amd back in the running on the desktop for gamers.

It depends a little. The current batch count which is what is used in games is around 15K for DirectX. In Mantle batches can go up to 40K without to much trouble.

If you checked some of the back story with Nitrous engine the software developers have stated that 80K or 100K should be available in next few years. That is a radical change in gaming development. That would however require a decent cpu to be able to process that amount of batches.

As it is now with Mantle the CPU (if it is very fast) just does nothing :). So with the batch count increasing there is still a need for 8 core CPU.
If you want to watch the video of Nitrous engine presentation on youtube just google towards the end in the video (last 15 minutes) they give out stats and analysis.
 
Last edited:
Back
Top