ambidextrous computing; AMD project skyla..skybridge!

[ShintaiDK]

The one thing I don't understand is this: if AMD claims the advantage of this approach is for example more flexibility for manufacture, e.g. they can release the same tablet with both Windows 8 on x86 and Android on ARM, that would still mean AMD has to produce both CPU designs in conjunction to stay within the same TDP limit.

To me that sounds like they will be hindering themselves with this, but maybe someone else is more knowledgeable on how feasible this is?

Its only purpose is cost reduction for AMDs design cost. One platform, one uncore, just replace x86 cores with ARM and wise versa. I wouldnt call it more flexibility for manufactors as such, but rather for AMD. The question is if this approach will end up hindering both concepts and end up with a product that fits nowhere.

 

[NTMBK]

Its only purpose is cost reduction for AMDs design cost. One platform, one uncore, just replace x86 cores with ARM and wise versa. I wouldnt call it more flexibility for manufactors as such, but rather for AMD. The question is if this approach will end up hindering both concepts and end up with a product that fits nowhere.

It certainly can reduce costs for OEMs too, so long as AMD produce a competitive ARM SoC. Design a single tablet motherboard design and chassis, and fit it out with either an x86 SoC for Windows or an ARM SoC for Android. Two very different products from one design.

(Yes, I know Android on x86 is perfectly feasible, but Intel seem to be keeping a tight leash on their port and AMD don't have the software engineering resources to do it by themselves.)

 

[mrmt]

It certainly can reduce costs for OEMs too, so long as AMD produce a competitive ARM SoC. Design a single tablet motherboard design and chassis, and fit it out with either an x86 SoC for Windows or an ARM SoC for Android. Two very different products from one design.

I don't know whether this move will have a meaningful impact on OEMs, because the TAM for both product lines may not be the same, but it's surely a win for AMD, because they should sell more products using the same platform, which dilutes the accrued OPEX in those units. It also helps to mitigate a weakness of their ARM line, which is the lack of scale. With this move they might get an scale greater than Nvidia. That should be enough to give them time to entrench themselves one or two niches. I wouldn't be optimistic for the Android market, because of the lack of LTE modem, but in embedded and servers, they should have a shot.

 

[Shakabutt]

So the thing that many of us are awaiting is the new CPU in 2015 ? or do we wait for the new thing in 2016, will it be competitive with anything from intel in terms of games etc ? or is amd done with desktop ? cant really make out anything from those slides

 

[ShintaiDK]

So the thing that many of us are awaiting is the new CPU in 2015 ? or do we wait for the new thing in 2016, will it be competitive with anything from intel in terms of games etc ? or is amd done with desktop ? cant really make out anything from those slides

No. Small cores is also their only future.

 

[TeknoBug]

"Skybridge"? Is this is a coincident with Intel's Skylake naming or is AMD being tryhards here?

 

[NTMBK]

No. Small cores is also their only future.

Wrong.

Since we're talking about a replacement for Bulldozer, this is an entirely different class of beast from the "Puma+" and Cortex-A57 cores in the first SkyBridge parts. AMD's execs noted that these are high-frequency, high-performance CPU cores that will span the range from laptops to desktops to servers—and not just "microservers." We don't know exactly how big they'll be compared to Bulldozer or, say, Haswell. The smartest play might be to aim for something a little smaller than those cores, but that's the basic class of performance they're undoubtedly meant to achieve. AMD is returning to its roots, aiming to produce a best-in-class big core.

 

[jdubs03]

No. Small cores is also their only future.

I disagree, Intel's competitive advantage sits within the technological capability of their fabs, and their "Core" line of processors. As long as they take core and scale down (like Broadwell-Y), they can maintain their advantage. I can't foresee them losing out on single-thread performance at any point in time.

I said it a while ago, but eventually Intel is going to merge Core and Atom into a new architecture. It would make sense if they could scale it all the way through, only needing one process, and focusing on one architecture instead of two. But then you're giving up the two points of differentiation in their product line.

Edit: I was thinking Intel not AMD. It seems as if AMD is focusing on those smaller cores, but I don't think they'll be done with bigger cores, there will still be areas where that extra performance is needed.

 

[rtsurfer]

No. Small cores is also their only future.

You would love that wouldn't you.?

Apparently you love vacations? Because you just earned one for a personal attack. Personal attacks will not be tolerated.
-ViRGE

 

[erunion]

Surprised? Foundry roadmaps and products is an entirely different thing.

I would say 14nm(read 20nm with FF) for 2016 is optimistic.

By 14nm not coming in 2015? No, EOY 2016 is a more reasonable time frame. (Though the slide makes no mention of 14nm, maybe it will be 2017 or later)

But 20nm in 2015 does surprise me. I was under the belief that GF 20nm was cancelled. So either its not or AMD is returning to TSMC

 

[ShintaiDK]

By 14nm not coming in 2015? No, EOY 2016 is a more reasonable time frame. (Though the slide makes no mention of 14nm, maybe it will be 2017 or later)

But 20nm in 2015 does surprise me. I was under the belief that GF 20nm was cancelled. So either its not or AMD is returning to TSMC

I would guess TSMC

 

[ShintaiDK]

Wrong.

You quote what is at best a guess. When you look on the R&D budget, there is no way AMD will be creating a big core that cost 4-5B$ or more to develop.

 

[NTMBK]

You quote what is at best a guess. When you look on the R&D budget, there is no way AMD will be creating a big core that cost 4-5B$ or more to develop.

I would say that AMD's statements about where their cores are targeted bears more weight than forum prognostication. Whether they succeed remains to be seen, but that is their stated target.

 

[AtenRa]

You quote what is at best a guess. When you look on the R&D budget, there is no way AMD will be creating a big core that cost 4-5B$ or more to develop.

What he quoted was spoken by Jim Keller himself. He precisely said they took the best of both Big-core(BD) and small core (Cat) architectures and combined them together to create the next x86 big core. They took the high-frequency and small core design and they are making a new x86 big-core around those characteristics.

 

[SViscusi]

Its only purpose is cost reduction for AMDs design cost. One platform, one uncore, just replace x86 cores with ARM and wise versa. I wouldnt call it more flexibility for manufactors as such, but rather for AMD. The question is if this approach will end up hindering both concepts and end up with a product that fits nowhere.

You say it's only "benefit is cost reduction" as if that's a bad thing or that only AMD itself will benefit from reduced costs. If a company is able to reduce the cost to manufacture a product it will benefit more than just that company. Customers will see it directly with more aggressive pricing throughout the life of the product. The market will see it indirectly through the savings generated from the cost reduction used to shore up a company's financial health and be reinvested into the company raising the quality of future parts. At a minimum, a more profitable company will provide more competition forcing it's competitors to produce better options which will benefit all customers.

How could any of that possibly be a bad thing?

 

[NostaSeronx]

The funny part is that Dozer/Big Cores and Cat/Small Cores is that they are already based on each other. I can see them fusing the design methodology to make a middle core that can work in HPC down to Essentials. I do not see in the foreseeable future seeing a High Performance Cat Family/Big core in the HPC market to mainstream markets.

The main reason that Bobcat and Jaguar core has a 2 ALU/2 AGU/2 FPU array per processor is to have parity with Dozer cores.

 

[erunion]

I would guess TSMC

Yeah, that seems to be the only plausible answer.

Its looking like 20nm gpus will come in 2015 as well. I wonder how this will play out, TSMC 20nm seems to be a troubled node.

 

[NostaSeronx]

20nm from GlobalFoundries and TSMC are currently to costly to be used in a high-medium volume capacity. Apparently, 20nm from GlobalFoundries will be the first to hit perfect cost/shrink metric. That is why all of AMD's design will be going from TSMC28 to GF28A(28SHP) then to GF20AN(20SHP).

 

[shady28]

ISA is mostly irrelevant out of the core microarchitecture. Of course there are system architecture specifications, but the potential and ease of wrapping them up is far greater than in a core microarchitecture. So if you design from sketch those parts with compliance of both the x86 and ARM system architectures, you can reuse them among all products, save you some costs and gain design flexibility. In this case, it is the on-chip SoC interconnect that chains up all the first-class citizens, last-level caches, I/O complexes, memory controllers and perhaps also system functions (e.g. interrupt controllers, debug bus, etc). Once you got this fabric right, plug and play of I/O and accelerator blocks is an easy job, as they are ISA agnostic by nature (e.g. abstracted behind PCIe).

So they are hedging their bets on x86 and ARM?

 

[erunion]

It certainly can reduce costs for OEMs too, so long as AMD produce a competitive ARM SoC. Design a single tablet motherboard design and chassis, and fit it out with either an x86 SoC for Windows or an ARM SoC for Android. Two very different products from one design.

(Yes, I know Android on x86 is perfectly feasible, but Intel seem to be keeping a tight leash on their port and AMD don't have the software engineering resources to do it by themselves.)

Is AMD even targeting tablets? They seem pretty vague about actual products.

 

[Idontcare]

Yeah, that seems to be the only plausible answer.

Its looking like 20nm gpus will come in 2015 as well. I wonder how this will play out, TSMC 20nm seems to be a troubled node.

Troubled node? Why do you say that?

 

[Topweasel]

The WiiU can already boot into Wii mode on the chip level. I don't see why it's so far fetched that this chip would do the same for ARM-x86. I think they are talking about an integrated ARM/x86 CPU. Just not one where the cores operate simulataneously.

Wii to Wii U, the architecture for the Wii and the WiiU is unchanged. In fact the WiiU is just a die shrunk Wii internally. But really for the most part the WiiU isn't architecturally much different then the Dolphin in the GC. The need to boot to Wii in the WiiU has more to do with OS compatibility with the games and not emulation. This is comparable to someone in 2001 having a dual boot PC, 1 partition with Windows 2000, and another partition for Windows 98SE for the older games that wouldn't run on 2000.

 

[mrmt]

I would guess TSMC

AMD cannot use TSMC.

 

[erunion]

Troubled node? Why do you say that?

Several minor things.

20nm is what caused nvidia's infamous wafer cost revolt.
http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless

Now the rumor is there will be no 20nm GPUs until 2015. And that Nvidia is back porting Maxwell to 28nm for its 800 series.

And unfortunately I don't have a link, but I remember someone from TSMC saying that in hindsight they probably shouldn't have launched 20nm planar and then transitioned to finfet a year later.
20nm was always going to be a short lived node, but it will be interesting to see if customers decide to skip 20nm and go straight to 16nm

 

[cbn]

I wouldn't be optimistic for the Android market, because of the lack of LTE modem, but in embedded and servers, they should have a shot.

Anyone want to field opinions on why AMD chose cortex A57 for their Heirofalcon storage server SOC.

I would have thought a cat core would have been a great fit for that one.

P.S. A few years ago, I would have thought AMD's first ARM SOC would have been an APU (ie, AMD's equivalent of Nvidia Denver) due to the general direction consumer software was going. But storage server? An ARM CPU core rather than x86 cat core? That is harder for me to understand.