ambidextrous computing; AMD project skyla..skybridge!

[Third_Eye]

Your end-2015 estimate for FinFET seems optimistic to me.

And Intel was supposed to have 14nm FinFET products by now right? It has canned its Chandler Arizona plant which was supposed to begin 14nm manufacturing and 450nm wafers.
So the process lead is reducing (I agree we have to wait till 2015 to see if TSMC as well as SMSG/GF can bring their equivalent node to the market). Manufacturing advantage (in the form of brand new fab) is no longer there.

That's like saying that FX-9590 can compete with i7. Yes it might be able to compete, but its TDP is twice as much.

We all can agree that the area that is "hot" for all chip manufacturers is how "cool" they can get their products run. That is precisely the reason 22nm Haswell provided very little performance improvements over Ivy but improved the performance/power markedly.
Similarly AMD too has been moving in that direction since Richland (which was better power Trinity at the same 32nm PD-SOI) as well with Steamroller now going into 28nm Bulk HKMG at 95W max TDP to Carrizo which will be 65W max TDP at the above node.

So now we have the following based on the below 2 leaks that WCCFTECH has posted.

http://wccftech.com/amd-mobile-kaveri-crystal-series-leaked/ - Models
http://wccftech.com/amd-kaveri-mobile-apus-vs-ulv-haswell-benchmarks/ - Comparisons

At the ULV category (19W), AMD is closing greatly on CPU performance (bettering i5 and i3 and possibly losing to i7), and extending the Graphics lead more.
We still have not factored in the future potential with HSA which would be a bonus.

And ULV too is a cash cow compared to desktops due to it ending in notebooks, 2 in 1s or in rare cases, even high-end tablets (like Surface Pro variants). Now AMD is competing in atleast some of them. And with 28nm product reaching 22nm performance at the same/similar power envelope.

Add to it we have Beema and Mullins going into the Baytrail territory with better performance/watt at the same 28nm but GF instead of TSMC.

So the 1+ node advantage + architectural advantage (Haswell vs the Dozer) + Single thread advantage is slowly being undercut in the x64 world.
Add to it AMD has a dedicated revenue stream from its console SOCs where it need not compete with INTC. So it can fight Intel in the x64 area with renewed vigor and that is happening.

Today an Apple dual core ARMv8 ISA compatible custom SOC @ 1.3GHz processor (purely for mobile) made at 28nm HKMG puts out equal or better performance than a quad x64 core Baytrail manufactured at 22nm FinFET under Geekbench running at higher clock speed. That shows that with the right design team, ARM can scale up and compete in the target market. It also shows that x86/x64 can scale down but the momentum is on the ARM side for now and INTC is being attacked (bitten) from all sides. As INTC tries to shut ARM from the latter's strong areas and losing money in the process, its traditional safe money making markets are being attacked/put to risk with another "ARM"y of Davids.

 

[~CS~]

didn't take long for this thread to be all about Intel...sigh.

..and zero attention from mods, I think that piesquared guy had a good point, too bad they banned him. :biggrin:


Mod callouts are not allowed.
Markfw900

 

[frozentundra123456]

..and zero attention from mods, I think that piesquared guy had a good point, too bad they banned him. :biggrin:

Yea, became more about bashing intel and predicting their demise in mobile. Dont think lack of attention from the mods to this thread shows any pro intel bias.



Mod callouts are not allowed.
Markfw900


Back to the topic, I think Intel has a very tough road in phones, a better chance in tablets. Regarding AMD though, I dont see how one thinks ARM will suddenly become their saviour, since they have to compete against the entrenched giants as well, and with much less resources than intel.

 

[blackened23]

..and zero attention from mods, I think that piesquared guy had a good point, too bad they banned him. :biggrin:

Almost reminds you of how the Bay Trail and Haswell Pentiums threads are littered with comparisons to AMD chips, doesn't it. It's almost as if............a competing product is relevant to discussion no matter how annoying it can be. Right? I dunno.



Mod callouts are not allowed.
Markfw900



It's a competing product to stuff that intel is offering or will be offering. Parallel conclusions will be drawn. I personally think a competing product is relevant to discussion. And if you think ARM will be the glorious savior of AMD, well, have fun with that. It's not like AMD won't be competing with qualcomm and the other giants, if you think that AMD will suddenly crush them on their first outing. Well. I'm 99.9% sure things won't happen that way. We'll see in....2016 though. Cross your fingers and such.

 

[shady28]

I feel compelled to respond with a simple "wow" type of awestruck compliment as your post so succinctly captures what is different about the present situation, and in all the right ways :thumbsup:

Intel is limiting their process node advantage to the innovations which their IC design teams can come up. That is only a win if your process folks AND your design folks are bar-none the world's best in every way possible...and the marketing/business/sales folks must be on top of their game as well.

But ARM is doing something that is very smart, leveraging diversification across IC design teams and companies to determine the path of least resistance in terms of defining for them what the market wants, and where next year's innovation is going to come from.

Intel must knock this out of the park in order to survive, whereas ARM merely needs just one of their multitude of customers to come up with the next iPhone and they have themselves at least another decade of revenue secured.

Thanks for the compliments!

The variety of capabilities ARM provides makes mobile SoCs incredibly interesting, and provides incredible variety and differentiation between products.

As you say Intel has to hit the ball out of the park in mobile to fight this. They would really need to make something so good that all (or at least most) of the variety and differentiation ARM partners provide is subsumed and becomes a non factor.

And perhaps back to the topic at hand -

If we acknowledge that the strength of ARM is how it empowers its partners to differentiate and cater to specific needs, how could that be used to affect ARM server platforms?

I am just thinking to myself, will we have A57 ARM cores with modules that somehow speed up common server applications like MySQL, or Apache? Or perhaps provide specialized management tools? I don't think this chip from AMD is it, but it would seem a logical strength of ARM to see a continuation of what we've seen in mobile creep into the server space.

 

[cbn]

<redacted>


thread crapping is not allowed here
Markfw900

 

[raghu78]

Bump ^^^^

28nm HKMG for SoFIA 3G or 28nm poly Si?

(Anyone want to wager a guess)

28 nm HKMG ( 28 HPM version) for SOFIA 3G / LTE. Its been confirmed

http://www.digitimes.com/news/a20140109PD215.html

 

[cbn]

28 nm HKMG ( 28 HPM version) for SOFIA 3G / LTE. Its been confirmed

http://www.digitimes.com/news/a20140109PD215.html

Thank you very much for that info.

 

[Idontcare]

If we acknowledge that the strength of ARM is how it empowers its partners to differentiate and cater to specific needs, how could that be used to affect ARM server platforms?

I am just thinking to myself, will we have A57 ARM cores with modules that somehow speed up common server applications like MySQL, or Apache? Or perhaps provide specialized management tools? I don't think this chip from AMD is it, but it would seem a logical strength of ARM to see a continuation of what we've seen in mobile creep into the server space.

I think that is exactly where you are going to see a significant percentage of the market exist in say 10 years or so.

Custom enough processors that don't carry the baggage of being a general purpose processor is going to deliver the best possible performance/watt in any given application. That has always been true.

What has made the "all-purpose" CPU advantageous is the cost angle. But the cost angle (a one-time hardware investment, capex) is rapidly being offset and outweighed by the energy-cost aspect embodied by TCO.

That spells opportunity.

 

[Vesku]

AMD has been advertising their design services. Can definitely see at least the potential but Intel apparently has done some of this in the past and even for big companies the upfront cost seems to be daunting. I'd imagine the costs will be less than Intel's but will it be low enough? There is also the issue that it adds another 1 or more participants to the business secrets risk pool, so that not only will the foundry get a peak at what you are implementing and its business implications but at least one company that can even more directly implement or utilize that knowledge (well if you use Samsung as foundry that's already a factor) and potentially more if you hired other companies to design the custom logic blocks.

There are also issues intertwined with costs. Confidence that the custom parts will be delivered on time and perform within expectations. An environment not conducive for quick and painless contract negotiations. AMD certainly has (had? - layoffs) respectable design experience but their own product launch recent history, last few years, won't be a great marketing tool.

Edit: On further pondering, perhaps AMD and Samsung will announce a design alliance to dovetail with GlobalFoundries adopting Samsung's fab process. If so, I'd expect an announcement by early 2015 pre-Skybridge.

Intel must have had a rethink on custom pricing in the last year or two - http://gigaom.com/2013/07/22/heres-why-intel-announced-a-custom-silicon-business-today/

If AMD and Samsung don't announce a design alliance I think that will be a business misstep.