How will AMD answer the challenge posed by Haswell?

[pelov]

As you point out with IA64, x86 already won. This also includes against ARM for the very same reasons. Plus a few more in that case.

It doesn't. ARM has gone 64-bit as well, so I don't see how that's analogous. The IA64 was a RISC architecture that attempted to flip around a market gunning for x86, whereas in this case, x86 entering mobile would be like IA64 entering the x86_64 bit world back then. It's the underdog rather than the dominant force. Obviously, the roles are reversed in the desktop and server space where it's still x86_64 that dominates.

There's no doubt that Intel knows how to handle RISC and CISC, and the hybrids we have today that we call x86 processors, but I don't see how any of that favors x86 in mobile. Today, Intel isn't as willing to make another RISC-like architecture for mobile when they've got a stranglehold on x86 that they didn't have back then. Remember that VIA wasn't a nonexistent force and AMD was actually quote potent. Now, it's just Intel with AMD slipping farther and farther behind. It's a completely different era of computing now. A majority of chips entering consumers hands are on the ARM side, not x86. Just a few years ago that would have been seen as blasphemous and utterly impossible. It's the same reason Intel hasn't yet developed a proper SoC yet that can challenge Qualcomm. They just didn't see it coming.

Obviously, they're far better prepared than AMD, who is selling us a 40nm 1ghz Brazos platform for tablets...

3. Modify the DDR3 bus to be able to share the bus with NAND. Provide a home for multichip multiplexing heterogenous memory (flash + DRAM) and they will come. This would of course also provide a compatible home for PCM when it arrives.

They really need this above all else. I do think HSA will do quite well with ARM simply because they'll take any compute power they possibly can to distance themselves from rivals, but unless AMD is able to quit relying so heavily on that DDR3 bus, they're screwed.

 

[ShintaiDK]

It doesn't. ARM has gone 64-bit as well, so I don't see how that's analogous. The IA64 was a RISC architecture that attempted to flip around a market gunning for x86, whereas in this case, x86 entering mobile would be like IA64 entering the x86_64 bit world back then. It's the underdog rather than the dominant force. Obviously, the roles are reversed in the desktop and server space where it's still x86_64 that dominates.

There's no doubt that Intel knows how to handle RISC and CISC, and the hybrids we have today that we call x86 processors, but I don't see how any of that favors x86 in mobile. Today, Intel isn't as willing to make another RISC-like architecture for mobile when they've got a stranglehold on x86 that they didn't have back then. Remember that VIA wasn't a nonexistent force and AMD was actually quote potent. Now, it's just Intel with AMD slipping farther and farther behind. It's a completely different era of computing now. A majority of chips entering consumers hands are on the ARM side, not x86. Just a few years ago that would have been seen as blasphemous and utterly impossible. It's the same reason Intel hasn't yet developed a proper SoC yet that can challenge Qualcomm. They just didn't see it coming.

Obviously, they're far better prepared than AMD, who is selling us a 40nm 1ghz Brazos platform for tablets...



They really need this above all else. I do think HSA will do quite well with ARM simply because they'll take any compute power they possibly can to distance themselves from rivals, but unless AMD is able to quit relying so heavily on that DDR3 bus, they're screwed.

I think you misunderstood why x86 won over IA64. And ARM aint 64bit yet.

A hint for why you can find in why almost everyone dumps WinRT for Win8 with tablets.

Or why 10 years later your latest game is still 32bit x86.

 

[pelov]

I think you misunderstood why x86 won over IA64

That had mostly to do with market trend and ease of use/development. Who besides Oracle has ever developed anything for IA64? or even wanted to?

And ARM aint 64bit yet.

Yes it is. They've also crossed over the out-of-order hurdle that many considered a CISC-only shindig.

 

[Nemesis 1]

Nemsis, I'm finding it difficult to follow you and I highly doubt I'm the only one. What I will leave you with is perhaps a heavy dose of reality:





The same TI that's pulling out their now defunct OMAP chips is ahead of Intel.

I do think AMD has a unique opportunity here. They're not as tied to x86 as Intel is, thus the flexibility that allows could perhaps present them a unique niche in the market where they can provide both, and maybe even both on the same chip. They're IP is quite weak though, and the companies making money in mobile are the ones with very good wireless/LTE IP -- something AMD lacks entirely.

On one side you've got the 500lb blue gorilla and on the other you've got cutthroat competition and a blistering pace and an ever-changing market, players and their respective positions. If they can net something in the middle they might provide something interesting, but I wouldn't count on a hybrid chip as a long term solution but rather a stopgap.

Ya I know I hard to follow. Intel just finely created an Atom that is good enough for smartphones. What is and what was and what will be . AS desktops and the like fade . The Smartphone power performance will matter . As form factor shrinks the amount of work in this space will be come more power and efficient in its productive state . Just because it matters not today whats in a smart phone . That doesn't mean it will remain thus .

 

[pelov]

The Smartphone power performance will matter . As form ffactor shrinks the amount of work in this space will be come more power and efficient in its productive state . Just because it matters not today whats in a smart phone . That doesn't mean it will remain thus .

...but it's not outperforming the ARM architectures.

I would agree with you if x86 absolutely blew ARM out of the water and Intel's Atom's, at the same TDP, were capable of much better performance. In fact, it's the opposite. The ARM architectures lost only a single benchmark while the same phones/models also outperformed it in battery life.

I think you're overestimating just what Intel is capable of. They're a monster in the desktop/server space, but as soon as you get into mobile, they're chips are bigger, more expensive, more power hungry, don't perform all that much better (and often worse), and they lack crucial implementations like LTE.

It's not a poor first attempt. Far from it. But to think they'll retake the crown when they've quite clearly got such a long road ahead of them? They can't exactly rely on Microsoft's dominance anymore either. That Wintel marriage is tearing at the seams.

 

[Nemesis 1]

I'm not claiming their brand value would hurt them, no- I agree most people don't care what's inside their phone. I was simply using those stats to disagree with nemesis' claim that Intel's brand power gave some sort of edge over Apple.

Well you could have assumed thats what I meant . But Intel needs to have its own line of smartphones and tablets . ON the HIGH END side only to deal with apple . Inte slapped Applle around befor with the aid of MS . It can happen again . Intel invested in Sharp for a reason. Intel also Did there own smartphone design thats selling today . When the 22nm chip arrives intel needs to brand the highend chip as Intell smartphone . Intel needs only look at Apple and samsung earnings

 

[ShintaiDK]

...but it's not outperforming the ARM architectures.

I just want to point out the different components in smartphones, just like laptops. Some ship with singlechannel memory and so on. In case of the XOLO, it ships with 400Mhz memory. The iPhone 5 for example ships with what, 800Mhz?

 

[pelov]

I just want to point out the different components in smartphones, just like laptops. Some ship with singlechannel memory and so on. In case of the XOLO, it ships with 400Mhz memory. The iPhone 5 for example ships with what, 800Mhz?

That was also Intel's reference design.

Keep in mind that the TDP is higher in Intel's reference design, thus the headroom in the iPhone allows for faster clocked memory.

You're painting it out to be an ARM-skewed test. If you want to blame someone, blame Intel for not strictly limiting it to 800mhz memory in their reference design.

And there's a reason it's in only 3(?) phones: nobody wants it. Do you think Samsung is lining up to order a few million Atoms for their smartphones and tablets? Apple? They've got their own SoCs brewing, while Qualcomm is snatching up all the rest. Until Intel has something of value to offer to them and decreases their price to a level where there are companies willing to take the risk of going x86 in smartphones (remember why the Itanium failed?), it won't happen. They need incentive to switch over. Currently, there are more applications, cheaper prices and better battery life all associated with ARM while Intel only has x86 to offer at higher prices. So what? Can you run Photoshop on your smartphone or tablet? Would you even want to?

 

[ShintaiDK]

That was also Intel's reference design.

Keep in mind that the TDP is higher in Intel's reference design, thus the headroom in the iPhone allows for faster clocked memory.

You're painting it out to be an ARM-skewed test. If you want to blame someone, blame Intel for not strictly limiting it to 800mhz memory in their reference design.

And there's a reason it's in only 3(?) phones: nobody wants it. Do you think Samsung is lining up to order a few million Atoms for their smartphones and tablets? Apple? They've got their own SoCs brewing, while Qualcomm is snatching up all the rest. Until Intel has something of value to offer to them and decreases their price to a level where there are companies willing to take the risk of going x86 in smartphones (remember why the Itanium failed?), it won't happen. They need incentive to switch over. Currently, there are more applications, cheaper prices and better battery life all associated with ARM while Intel only has x86 to offer. So what?

Whats the price of the XOLO? Even tho Intel is getting their huge margins?

Nobody wants it? Are you even serious? XOLO in India, K800 in China. Orange sells its own branded one in France and the UK. And you could go on.

Oh, and some of those ARM chips are quadcores aint they?

 

[Exophase]

Intel got other uarchs besides x86, and developed quite a few non x86 chips. AMD developed what again?

The 29k: http://en.wikipedia.org/wiki/AMD_Am29000

And if you want to include it, the GPU side of things is becoming more and more general purpose-friendly. On that note, I would also consider Intel's Gen GPU architecture to be the only place where they're seriously pushing a pure non-x86, although LRB at least gets points for doing its own thing with the SIMD. It's hard to call them very serious about Itanium.

I don't think Intel was ever that devoted to XScale either. Their ARM license and development team was won out of a settlement from DEC whom they grabbed StrongARM from. Settlement, not purchase, as in they took what they could get and not necessarily something they highly valued. I fully expect that DEC was already underway developing XScale, and it wasn't that long before Intel was dumping it on the nearest buyer. That's ancient history now, but I'd have barely called it Intel's initiative in the first place.

You definitely shouldn't take home a message that Intel invested in ARM until they realized it sucked and were forced to go with x86. There are a reasons why x86 benefits Intel, but they're quite Intel-centric. By the early 2000s they were probably already coy about investing in other ISAs.

It doesn't. ARM has gone 64-bit as well, so I don't see how that's analogous.

I'm something of an ARM proponent but I'm first to admit that when it comes to 64-bit they're at a disadvantage. It takes a long time to go from an arch spec to shipping processors. Intel has been keeping 64-bit out of Medfield for some reason but they only need flip a switch on their next SKU to rectify that. For all their other processors, and for AMD's, they're already there. So if the market demands 64-bit in tablets in a year or two it could be a problem for ARM, although I suspect Apple and nVidia, and maybe Qualcomm all have a leg up with their own designs.

I just want to point out the different components in smartphones, just like laptops. Some ship with singlechannel memory and so on. In case of the XOLO, it ships with 400Mhz memory. The iPhone 5 for example ships with what, 800Mhz?

Where are you getting this information?

Memory ratings are notorious for being inconsistent. These days most people use "MHz" to mean "MT/s" (mega-transfers per second) which is twice the clock rate in DDR memory. But some people don't, leading to some bad comparisons. I'd eat my hat if a Medfield phone shipping today came with 200MHz/400MT/s LPDDR2 - is that even being sold? A more believable story would be that it's single channel. But if they did that I'd bet anything it was to save power and not costs, just like I'd bet that for the puny SGX540 Medfield is currently packing. No way is Intel going to cripple their reference platform unless they have to or they're idiots.

For what it's worth, iPhone 5 uses 533MHz/1066MT/s dual-channel LPDDR2.

 

[pelov]

Whats the price of the XOLO? Even tho Intel is getting their huge margins?

Nobody wants it? Are you even serious?

$420, though that's in India and that was in April.

Beginning April 23rd, Intel, through Lava International, began selling the Xolo X900 smartphone in India for INR 22000 (~$420 USD). As we’ve stated before, the design and construction of the Xolo X900 almost identically mirrors the Intel FFRD we’ve seen before, from the specifications and Medfield platform itself, to industrial design and exterior buttons.

http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone

Smaller battery and cheaper screen than both of the Samsungs yet pretty much neck-and-neck with respect to performance.

 

[ShintaiDK]

$420, though that's in India and that was in April.



http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone

Smaller battery and cheaper screen than both of the Samsungs yet pretty much neck-and-neck with respect to performance.

And what do those faster phones cost?

Diesize of the Z2460 is also much smaller than the competition.

 

[Ajay]

I think you misunderstood why x86 won over IA64. And ARM aint 64bit yet.

A hint for why you can find in why almost everyone dumps WinRT for Win8 with tablets.

Or why 10 years later your latest game is still 32bit x86.

The Golden Handcuffs, though M$ may have misplayed it's hand with Win8, so we'll see.

=
Yes it is. They've also crossed over the out-of-order hurdle that many considered a CISC-only shindig.

The architecture is here, but I don't think there are any commercial implementations yet. Several are due sometime this year though.

 

[ShintaiDK]

Where are you getting this information?

Just look under the reviews of phones.

http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone

 

[Nemesis 1]

...but it's not outperforming the ARM architectures.

I would agree with you if x86 absolutely blew ARM out of the water and Intel's Atom's, at the same TDP, were capable of much better performance. In fact, it's the opposite. The ARM architectures lost only a single benchmark while the same phones/models also outperformed it in battery life.

I think you're overestimating just what Intel is capable of. They're a monster in the desktop/server space, but as soon as you get into mobile, they're chips are bigger, more expensive, more power hungry, don't perform all that much better (and often worse), and they lack crucial implementations like LTE.

It's not a poor first attempt. Far from it. But to think they'll retake the crown when they've quite clearly got such a long road ahead of them? They can't exactly rely on Microsoft's dominance anymore either. That Wintel marriage is tearing at the seams.

Again I will say it . Intel just got into the smartphone. Lets wait for second generation intel phone chip. I mean christ Intel can run a haswell@8watts . This chip thats released in 2013 won't be a follower at all it will be best in class

 

[pelov]

Again I will say it . Intel just got into the smartphone. Lets wait for second generation intel phone chip. I mean christ Intel can run a haswell@8watts . This chip thats released in 2013 won't be a follower at all it will be best in class

That's what they need, but I don't understand why some of you, including yourself, are blindly accepting that it will happen?

Brian summed it up really well:

On the one hand it's a good thing that you can't tell an Intel smartphone apart from one running an ARM based SoC, on the other hand it does nothing to actually sell the Intel experience. Intel is never taken seriously in markets where it relies on being good enough, and it moves mountains in those where it's the best. That's what Intel needs to really build credibility in the smartphone space. A little was earned by getting this far, but its reputation will be made based on what happens next. There's obviously a strategy here, but I'm curious to see it unfold. Intel can be a fierce competitor in any space where it feels threatened. What I'm waiting for is that Conroe moment, but in a smartphone.

And claiming the chip is smaller while the cellular functions are still outside of the chip (built at TSMC), isn't something to brag about. They're on 32nm with Medfield while the competition that bests it is sitting at 45nm. Smaller? Yes. Better performing and better perf-per-watt with lower power draw? No. I'm not so sure that fab advantage will translate into more potent fabs in the mobile space, either.

 

[Exophase]

Just look under the reviews of phones.

http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone

So in other words, it's ambiguous. Note that it doesn't actually say 800MHz for the iPhone 4S, just "LPDDR2-800." Which is a real rating number. But not a clock speed.

Maybe you should ask Brian if he can clarify it. But I'm going to stand by my sentiments that putting 400MT/s (200MHz real clock) LPDDR2 in a recent phone would be very unusual. 800MT/s, 400MHz and higher LPDDR2 has been used in devices for nearly two years now.

 

[Nemesis 1]

That's what they need, but I don't understand why some of you, including yourself, are blindly accepting that it will happen?

Brian summed it up really well:



And claiming the chip is smaller while the cellular functions are still outside of the chip (built at TSMC), isn't something to brag about. They're on 32nm with Medfield while the competition that bests it is sitting at 45nm. Smaller? Yes. Better performing and better perf-per-watt with lower power draw? No. I'm not so sure that fab advantage will translate into more potent fabs in the mobile space, either.

Its OoO thats a big deal. Looking at Haswell 8 watt chip 10 watt if you preferr. I see intel has crossed the channel and its open calm sailing ffrom here on out . Intel other than samgsun have the luxury of best in class fabs. Why would you believe intel will stumble. Were intel screwed up was 5 years ago when they committed to the preset arch for 5 years . That was a bone head move and I let them know it to.

 

[Nemesis 1]

That's what they need, but I don't understand why some of you, including yourself, are blindly accepting that it will happen?

Brian summed it up really well:



And claiming the chip is smaller while the cellular functions are still outside of the chip (built at TSMC), isn't something to brag about. They're on 32nm with Medfield while the competition that bests it is sitting at 45nm. Smaller? Yes. Better performing and better perf-per-watt with lower power draw? No. I'm not so sure that fab advantage will translate into more potent fabs in the mobile space, either.

Its OoO thats a big deal. Looking at Haswell 8 watt chip 10 watt if you preferr. I see intel has crossed the channel and its open calm sailing from here on out . Intel other than samgsun have the luxury of best in class fabs. Why would you believe intel will stumble. Were intel screwed up was 5 years ago when they committed to the preset arch for 5 years . That was a bone head move and I let them know it to. I like intel naming . Sandy bridge Ivy Bridge both indacating a crossing. Knights ferry . showing another crossing . knights corner showing domain.
haswell broadwell penwell . Not sure how I got 2 post here

 

[Nemesis 1]

You people keep looking at intel like their under constrainmment . Intel is not a monopoly in this space . I want to see intel slash m,argins for a time .. Knock some players out . and they can its legal in this space so long as they show a proffit their hands are free to play like they haven't been allowed for along time . Should be fun to watch . and needlle those that think arm will win LOL

 

[Blitzvogel]

Even if AMD did everything I suggested, they would still have a tough time. To recap:

1. Move one or more GPU SIMDS into each cpu FPU core. Rewrite the drivers from the ground up to accommodate and optimize.

Sounds like a good idea, but where would the TMUs and ROPs end up (I assume TMUs attached to the SIMDs), and how would the entire performance and power draw of the chip be changed as a result in making it all work in the first place? While I almost want to see this way of doing things as the future, I have a feeling AMD wanted to do this like Intel with Larrabee but saw some major drawbacks. Then would such a set up be able to manage and feed a high number of SIMDs per core in order to keep the per core x86 throughput relatively wide, unlike with Larrabee being built around dozens of cores?

Fusion is certainly AMDs baby now, and if Jaguar is decent in x86 and graphics performance while extremely power efficient, AMD may have another Brazos/Zacate like success on their hands, especially if they can get it moving at 1 GHz or so while still coming in under 5W. The quad-core configuration could actually be quite a performer, maybe equalling Trinity at similar low clocks in general performance.

 

[NTMBK]

Its OoO thats a big deal.

Medfield is most certainly not an OoO chip. The only OoO tablet chip currently out is Hondo.

 

[Idontcare]

Medfield is most certainly not an OoO chip. The only OoO tablet chip currently out is Hondo.

I think he is thinking of the 14nm Atom which is purported to be a complete microarchitectural redesign and will include OoO.

 

[Nemesis 1]

No . IDC I am not . I meant the next 22nm Atom for phones is suppose to OoO./ Pretty sure I read that in AT article. But its good that you brought it up . I best go check . Its easy to be wrong.

 

[NTMBK]

No . IDC I am not . I meant the next 22nm Atom for phones is suppose to OoO./ Pretty sure I read that in AT article. But its good that you brought it up . I best go check . Its easy to be wrong.

Yeah, 22nm Valleyview is going to be OoO. I got confused by your post, I thought you were saying that the current Atom is OoO - it wasn't the easiest post to follow, in all honesty