5 ms·

ARM Unfazed By Intel's New '3D' Chips

8 points by adeelarshad82 15y ago

pjscott 15y ago
This is what you would expect them to say, whether or not they're actually fazed by Intel's snazzy new 22nm process.
luu 15y ago
The ARM ecosystem innovates along multiple points, not only the process but also through micro‐architecture and system‐on‐chip design that enables optimized efficiency for a broad range of products I don't understand how this is even a response to Intel's announcement. Intel also "innovates along multiple points" "through micro‐architecture and system‐on‐chip design". Additionally, they also have the world's best process technology. They certainly publish about more interesting micro-architectural innovations, and I'd argue that their chips are more 'interesting' on a micro-architectural level. They're obviously behind on the SoC side, because SoCs haven't been a focus for them, but it's conceivable that they could produce better SoCs than ARM implementers do, if they put their mind to it. However, it's inconceivable that anyone on the planet will have a better process in the foreseeable future. The press release talks about working with IBM, but that's not very reassuring. I remember looking at IBM's process parameters and specs a few years ago, when they had press releases crowing about SOI, and Intel's nth generation bulk process was nearly as good as IBM's n+1 generation SOI process (and no doubt much cheaper, since it didn’t have to deal with all of the complexities of SOI), and Intel was about half a process generation ahead at the time. I'm not a fab guy, so I don't keep up on this stuff, but I hear Intel's now more than a full generation ahead, and that their advantage within a generation is larger than ever.
- ChuckMcM 15y ago
  Its a response because they had to respond, they were called out on the playground by Intel. I find it hugely amusing that Intel bought the StrongARM business from DEC (as well as the Alpha team back in the day) and put their 'embedded' guys on it (think 8051/8048) and were surprised that they didn't develop any market share. At the time Intel felt like 'embedded' chips had to be cheap, low power, and peripheral rich. They knew this to be true because they had tried an 8086 SOC chip, it was called the 80186 and it was an utter failure for them. (disclaimer I was the Systems Validation engineer on the 80186 when I worked there in '84). So the company learned that "nobody" wanted high end SOCs, what they wanted was a replacable "core" CPU, an IO chip which provided a variety of peripherals, and a memory controller chip that could interface to the current memory technology. Their reasoning was quite sound, and literally for decades they have been executing against this 'north bridge' / cpu / 'south bridge' architecture plan. AMD through a curve ball when they integrated the memory controller on the Opteron, it made for faster memory performance and they got some design wins out of it, however Intel knew that memory tech would change and AMD would be up a creek, which they sorta were when FBDIMMs and DDR3 started coming out and their chips only talked DDR2. The scenery changed however with an explosion of embedded systems. Did you know that MIPS was the #1 embedded computer architecture (by volume) for a while because every HP printer had one in it? SyByte (acquired by Broadcom) answered with a nice 64 bit embedded MIPS architecture chip and it fueled a lot of embedded devices with a lot of potential. And of course ARM came out of 'nowhere' (not really of course, its been 35 years of hill climb but it seemed to many that they were suddenly more relevant) and they started getting design wins in cell phones which grew to smart phones and then to tablets. There was a great big 'arm' - ageddon when a ASUS and a couple of other laptop makers were going to ship ARM based laptops and (rumor has it) Intel and Microsoft threatened withdrawing all support if they shipped them. (which would have killed these companies outright) But there really isn't anything those two can do about Apple and Apple's use of ARM in the iPod and iPhone and then iPad really created a volume channel for ARM which, given the resources of more licenses has been evolving the architecture more rapidly, and Intel sold StrongARM to Marvell who has been using it in GuruPlug and elsewhere so it suddenly becomes a 'threat' on Intel's radar when they find the fastest growing segment of electronics that need high margin microprocessors (the place they like to lead) isn't using their architecture. So to anyone who has listened to their marketing spin on how you should use an Atom instead of an ARM, and told Intel "Except it still takes 4 chips for just the base system and its slower and it takes more power and it doesn't use your 64 bit extensions, ..." Intel needs to respond, because they know they are losing the next design to ARM, again. And Intel gets lower margin on Atom than it does on Pentium * so it has to be sensitive about making Atom "too" good. Rock meet hard place, and there you will find Intel. Which brings us to the current round of chip wars. In ARM's camp their biggest issue is that they don't have a 64 bit ISA specification in production [1] (although I believe the Tegra3 is supposed to be 64 bit) Intel's biggest problem is that they don't have anywhere close to the power headroom that ARM does. The 'next round' (the iPad 3, the Android Tablets, smartphones) seems to be dominated by ARM architectures and Intel wants in apparently. This bit about their 3D architecture is a 'talking point' to explain why things will be different now than they were the last time they tried to get folks to design in an x86 ISA. [1] http://www.arm.com/products/processors/index.php http://www.arm.com/products/processors/index.php
barrkel 15y ago
As the investment levels required to keep up with fab technology get higher and higher, the more it starts looking like a natural monopoly. I suspect we'll be relying on better programming techniques for increasingly parallel machines to save us from the fate of excessive dependence on Intel.
stcredzero 15y ago
As Intel begins to advance its line of Atom processors on its best available manufacturing processes over the next few years, it could begin to compete with ARM much more readily. Apple could change its toolchain so that iOS apps are compiled as ARM/Atom "universal" binaries. Apple could then make both ARM and Atom devices. He also said that ARM is testing 20nm and 22nm chips with partners and is collaborating with IBM on designing 14nm chips, a process node that is still several years away from high-volume manufacturing. Tri-Gate gives an advantage within the same process generation. While ARM's RISC architecture does give a fundamental advantage in reduced instruction decoding overhead, this overhead can be seen as more or less a constant factor, which diminishes as fabrication technology improves.
goalieca 15y ago
I suppose it would be possible to licence both ARM and intel to make a really sweet chip.
- pjscott 15y ago
  Do you mean creating an ARM chip and manufacturing it in Intel's fabs? If Intel would consent to it, and if the design could be brought to market fast enough, that could be a killer combination for the company that manages to pull it off.
nate_meurer 15y ago
I'm confused. I thought ARM was just an architecture, completely unbound to any process technology. Is it not plausible that ARM dies will eventually be fabbed on tri-gate as well? This article also makes it sound like Intel has a monopoly on 3D, but I believe I read elsewhere others are looking at it too, notably TSMC.
MaysonL 15y ago
Hey, if Intel wanted to, they could buy an ARM license, and build ARM processors using this process. Instruction set architecture and chip process technology are largely orthogonal.

8 points by adeelarshad82 15y ago

This is what you would expect them to say, whether or not they're actually fazed by Intel's snazzy new 22nm process.

luu 15y ago

The ARM ecosystem innovates along multiple points, not only the process but also through micro‐architecture and system‐on‐chip design that enables optimized efficiency for a broad range of products I don't understand how this is even a response to Intel's announcement. Intel also "innovates along multiple points" "through micro‐architecture and system‐on‐chip design". Additionally, they also have the world's best process technology. They certainly publish about more interesting micro-architectural innovations, and I'd argue that their chips are more 'interesting' on a micro-architectural level. They're obviously behind on the SoC side, because SoCs haven't been a focus for them, but it's conceivable that they could produce better SoCs than ARM implementers do, if they put their mind to it. However, it's inconceivable that anyone on the planet will have a better process in the foreseeable future. The press release talks about working with IBM, but that's not very reassuring. I remember looking at IBM's process parameters and specs a few years ago, when they had press releases crowing about SOI, and Intel's nth generation bulk process was nearly as good as IBM's n+1 generation SOI process (and no doubt much cheaper, since it didn’t have to deal with all of the complexities of SOI), and Intel was about half a process generation ahead at the time. I'm not a fab guy, so I don't keep up on this stuff, but I hear Intel's now more than a full generation ahead, and that their advantage within a generation is larger than ever.

ChuckMcM 15y ago
Its a response because they had to respond, they were called out on the playground by Intel. I find it hugely amusing that Intel bought the StrongARM business from DEC (as well as the Alpha team back in the day) and put their 'embedded' guys on it (think 8051/8048) and were surprised that they didn't develop any market share. At the time Intel felt like 'embedded' chips had to be cheap, low power, and peripheral rich. They knew this to be true because they had tried an 8086 SOC chip, it was called the 80186 and it was an utter failure for them. (disclaimer I was the Systems Validation engineer on the 80186 when I worked there in '84). So the company learned that "nobody" wanted high end SOCs, what they wanted was a replacable "core" CPU, an IO chip which provided a variety of peripherals, and a memory controller chip that could interface to the current memory technology. Their reasoning was quite sound, and literally for decades they have been executing against this 'north bridge' / cpu / 'south bridge' architecture plan. AMD through a curve ball when they integrated the memory controller on the Opteron, it made for faster memory performance and they got some design wins out of it, however Intel knew that memory tech would change and AMD would be up a creek, which they sorta were when FBDIMMs and DDR3 started coming out and their chips only talked DDR2. The scenery changed however with an explosion of embedded systems. Did you know that MIPS was the #1 embedded computer architecture (by volume) for a while because every HP printer had one in it? SyByte (acquired by Broadcom) answered with a nice 64 bit embedded MIPS architecture chip and it fueled a lot of embedded devices with a lot of potential. And of course ARM came out of 'nowhere' (not really of course, its been 35 years of hill climb but it seemed to many that they were suddenly more relevant) and they started getting design wins in cell phones which grew to smart phones and then to tablets. There was a great big 'arm' - ageddon when a ASUS and a couple of other laptop makers were going to ship ARM based laptops and (rumor has it) Intel and Microsoft threatened withdrawing all support if they shipped them. (which would have killed these companies outright) But there really isn't anything those two can do about Apple and Apple's use of ARM in the iPod and iPhone and then iPad really created a volume channel for ARM which, given the resources of more licenses has been evolving the architecture more rapidly, and Intel sold StrongARM to Marvell who has been using it in GuruPlug and elsewhere so it suddenly becomes a 'threat' on Intel's radar when they find the fastest growing segment of electronics that need high margin microprocessors (the place they like to lead) isn't using their architecture. So to anyone who has listened to their marketing spin on how you should use an Atom instead of an ARM, and told Intel "Except it still takes 4 chips for just the base system and its slower and it takes more power and it doesn't use your 64 bit extensions, ..." Intel needs to respond, because they know they are losing the next design to ARM, again. And Intel gets lower margin on Atom than it does on Pentium * so it has to be sensitive about making Atom "too" good. Rock meet hard place, and there you will find Intel. Which brings us to the current round of chip wars. In ARM's camp their biggest issue is that they don't have a 64 bit ISA specification in production [1] (although I believe the Tegra3 is supposed to be 64 bit) Intel's biggest problem is that they don't have anywhere close to the power headroom that ARM does. The 'next round' (the iPad 3, the Android Tablets, smartphones) seems to be dominated by ARM architectures and Intel wants in apparently. This bit about their 3D architecture is a 'talking point' to explain why things will be different now than they were the last time they tried to get folks to design in an x86 ISA. [1] http://www.arm.com/products/processors/index.php http://www.arm.com/products/processors/index.php

barrkel 15y ago

As the investment levels required to keep up with fab technology get higher and higher, the more it starts looking like a natural monopoly. I suspect we'll be relying on better programming techniques for increasingly parallel machines to save us from the fate of excessive dependence on Intel.

stcredzero 15y ago

As Intel begins to advance its line of Atom processors on its best available manufacturing processes over the next few years, it could begin to compete with ARM much more readily. Apple could change its toolchain so that iOS apps are compiled as ARM/Atom "universal" binaries. Apple could then make both ARM and Atom devices. He also said that ARM is testing 20nm and 22nm chips with partners and is collaborating with IBM on designing 14nm chips, a process node that is still several years away from high-volume manufacturing. Tri-Gate gives an advantage within the same process generation. While ARM's RISC architecture does give a fundamental advantage in reduced instruction decoding overhead, this overhead can be seen as more or less a constant factor, which diminishes as fabrication technology improves.

goalieca 15y ago

I suppose it would be possible to licence both ARM and intel to make a really sweet chip.

pjscott 15y ago
Do you mean creating an ARM chip and manufacturing it in Intel's fabs? If Intel would consent to it, and if the design could be brought to market fast enough, that could be a killer combination for the company that manages to pull it off.

nate_meurer 15y ago

I'm confused. I thought ARM was just an architecture, completely unbound to any process technology. Is it not plausible that ARM dies will eventually be fabbed on tri-gate as well? This article also makes it sound like Intel has a monopoly on 3D, but I believe I read elsewhere others are looking at it too, notably TSMC.

MaysonL 15y ago

Hey, if Intel wanted to, they could buy an ARM license, and build ARM processors using this process. Instruction set architecture and chip process technology are largely orthogonal.