Can ARM stand up against rivals Intel and AMD in the server market?
Wrong question. If they're pushing into servers, I'd expect ARM to worry more about two things:
- Sun/Oracle Sparc doing DB/web server loads at low power (with IBM Power playing the high-VM-capacity game).
- It's not so much the core CPU/ISA, which ends up being a small portion of the chips. It's the expertise in multi-core buses, caches, and all related paraphernalia. Not exactly where ARM have been playing ...
Might be that ARM will be happy to let Google/Agnilux work on that, and cooperate on CPU tweaks, keeping the license revenue.
I think they are considering doing to x86 servers what x86 servers did to RISC servers. They grab the low-end, get volume and become more competitive. x86 competes by going further in the same direction it is now, which is by becoming faster. Like RISC, they will ultimately be reduced to a niche if they go this route.
And that would make me happy. I never liked the 8008 and its offspring. The ARM is much more elegant.
I think things have changed since clayton christensen wrote his book on disruptive innovation. Intel is very much aware of disruptive innovation, and is really trying to compete on the low end, see the celeron processor ,and the efforts intel is putting into lower cost, lower power atom processors .
Clayton christensen, in his book , recommended to assign disruptive innovation projects, to seperate business units.I remember that the celeron development was one of the highest achievement of the israel intel branch. so maybe they implmented christensen's advice.
IIRC, it was the Pentium M that came from the Israel branch. The first Celerons (again, IIRC) were common Pentium IIs without cache in the processor cartridge.
"What Intel did was set up a separate business unit in Israel down at the low end. They have a product called the Celeron chip which is much cheaper and came in at the low end of the market. "
http://www.cse.iitb.ac.in/~cs671/interview.doc
Kudos for the theme of your analysis, but is Intel really serious enough about the Atom? I haven't looked in quite a while, but when I did one of the big problems was that the chipsets they provided were not optimized for low power and consumed more than the processor itself. Have they corrected this, especially for the mobile market?
Then there's the question I have WRT to it being a horse race; as I understand it:
The Atom is in-order superscalar, 64 bit (all recent versions), faster than anything from ARM and consumes more power.
The Cortex A9 is out-of-order, 32 bit, slower than the Atom (don't know how that might play out with ARM 2-4 cores) and of course consumes less power.
I don't know about the chipset issue, and yes this is an important point.
But the biggest move that points to the seriousness of their approach , is moving to selling ATOM cores at TSMC.
selling cores is a low margin business, but it makes sense to enable fast time to market , and high integration, which are important for consumer products.
That of course that doesn't mean that intel would win, but i think they seriously try.
And on the third hand, much of the commentary I found on the net suggests that Intel has a disruptive technology problem here. Until they have a few notable successes that may be the way to bet, however hard they are trying.
except for tsmc , intel also works hard on developing it's own SOC:
"LG Unveils GW990 Smartphone with Intel's Moorestown Platform " jan2010.
so intel has one design win for it's platform. Also they have moorsetown-w which supports windows.i wonder what would a windows capable phone would look like.
> I never liked the 8008 and its offspring. The ARM is much more elegant.
I agree very much with you, even the 68K architecture was clean in comparison, and then there is MIPS.
But it's been a long time ago when the underlying CPU was visible enough for me to worry about its architecture, the only way I'm aware of it nowadays is by what kind of toolchain I have to use and a few optimization flags.
It may be buried under layer upon layer of runtime code, but I know it's there.
Oh boy... Now I'm getting depressed I have to use a PC... Just thinking it probably has a complete IBM 5150 in its chipset makes me want to wash my hands.
> It's not so much the core CPU/ISA, which ends up being a small portion of the chips. It's the expertise in multi-core buses, caches, and all related paraphernalia. Not exactly where ARM have been playing ...
ARM have been playing in this space for many years now. SoC fabric, such as the AMBA bus design, is a central part of the ARM offering.
Thanks for the point. I'm aware of the SoC issues in general - analog bits, etc. I meant high-performance items like 6MB caches, buses like the EV one in Alpha chips (with AMD and Intel descendents), etc. That was my doubt.
I am a huge ARM fan (am currently building a handheld based on the nvidia tegra developer devkit). However I feel sceptical that ARM will go anywhere on servers. Just like VIA didnt go anywhere when it was the only mainstream SOC around, just like ARM didnt go anywhere in netbooks and just like AMD has yes to penetrate significant market share inspite having a huge performance lead for ~2 years in the pentium 4 days. Intel is the big gorilla in the room and are very adept at coming up with solutions that get 80% of the way there (like atom) whenever a new market niche opens up. Then they use their marketing muscle to make sure that the new niche doesnt get exploited by an upcoming competitor. So ARM might make the best performance/watt chips out there, but if they dont have the muscle to get design wins and ship in actual servers then sadly they will continue to remain a fringe player in the server market.
so i absolutly agree, im also a big ARM fan,and id love to see them get into the server market. its about time for another architecture to be doin something.. i think with the right approach enough software "muscle" will get ported. the the proposition to have chips that perform at a tiny fraction of the power consumption as current chips is just too lucrative to give up.
I think you start to see ARM replacing intel. in some sense IPAD is replacing some netbooks.
ARM's biggest problem was windows on the netbook. usally when disruption happens , it replaces a whole value chain with another. that's what's the IPAD is doing, and chrome os might do.
So you would have the apple's and google's marketing machines (and other big companies around the ARM architecture), fighting against intel's. that's a more fair fight than intel VS amd.
I get this for utility servers, but can ARM chips address more than 4GB of memory? I currently use a couple of ITX Via boards for some items (e.g. gateway, name server), but I am wondering how far I could go with ARM.
No, it's still a 32 bit processor, with no signs of a 64 bit version (the last time I asked about this, someone said he'd found an item from 2001 promising a 64 bit version RSN).
They may have a chicken and egg problem here: with so much money being made in their current markets they might not be able to justify (to themselves, at least) the significant expense of a competitive 64 bit design. Then again, maybe with today's techniques it wouldn't be so hard to simply widen the Cortex A9, which is already up to 4-way SMP and out-of-order superscalar. Achieving those two features wasn't cheap, I'm sure.
though thats pretty much the only mention a quick search coughs up, so I wouldn't call it definitive, for one thing I don't know if the 1176 is in production of not.
edit: the ARM website doesn't seem to mention 64 bit anywhere either so toshiba may be engaging in a rather creative definition of 64bit.
28 comments
[ 3.9 ms ] story [ 79.0 ms ] threadWrong question. If they're pushing into servers, I'd expect ARM to worry more about two things:
- Sun/Oracle Sparc doing DB/web server loads at low power (with IBM Power playing the high-VM-capacity game).
- It's not so much the core CPU/ISA, which ends up being a small portion of the chips. It's the expertise in multi-core buses, caches, and all related paraphernalia. Not exactly where ARM have been playing ...
Might be that ARM will be happy to let Google/Agnilux work on that, and cooperate on CPU tweaks, keeping the license revenue.
And that would make me happy. I never liked the 8008 and its offspring. The ARM is much more elegant.
I doubt many Intel execs are willing to sacrifice their short-term bonuses for the survival of the company after they left or retired.
Then there's the question I have WRT to it being a horse race; as I understand it:
The Atom is in-order superscalar, 64 bit (all recent versions), faster than anything from ARM and consumes more power.
The Cortex A9 is out-of-order, 32 bit, slower than the Atom (don't know how that might play out with ARM 2-4 cores) and of course consumes less power.
On a speed/power basis, ARM still wins.
But the biggest move that points to the seriousness of their approach , is moving to selling ATOM cores at TSMC. selling cores is a low margin business, but it makes sense to enable fast time to market , and high integration, which are important for consumer products.
That of course that doesn't mean that intel would win, but i think they seriously try.
OK, absolutely yes, Intel is deadly serious about this market given that they did that.
On the other hand it looks like they've got a customer demand problem with this attempt: http://bits.blogs.nytimes.com/2010/02/24/a-tie-up-between-in...; the partnership is on "hiatus" as of the end of February but Intel hasn't given up.
And on the third hand, much of the commentary I found on the net suggests that Intel has a disruptive technology problem here. Until they have a few notable successes that may be the way to bet, however hard they are trying.
so intel has one design win for it's platform. Also they have moorsetown-w which supports windows.i wonder what would a windows capable phone would look like.
I agree very much with you, even the 68K architecture was clean in comparison, and then there is MIPS.
But it's been a long time ago when the underlying CPU was visible enough for me to worry about its architecture, the only way I'm aware of it nowadays is by what kind of toolchain I have to use and a few optimization flags.
Oh boy... Now I'm getting depressed I have to use a PC... Just thinking it probably has a complete IBM 5150 in its chipset makes me want to wash my hands.
ARM have been playing in this space for many years now. SoC fabric, such as the AMBA bus design, is a central part of the ARM offering.
(self-note: http://en.wikipedia.org/wiki/Advanced_Microcontroller_Bus_Ar...)
Unfortunately, yes, it might happen. Or it may be declared to be against the terms of service or something stupid like that.
ARM's biggest problem was windows on the netbook. usally when disruption happens , it replaces a whole value chain with another. that's what's the IPAD is doing, and chrome os might do.
So you would have the apple's and google's marketing machines (and other big companies around the ARM architecture), fighting against intel's. that's a more fair fight than intel VS amd.
They may have a chicken and egg problem here: with so much money being made in their current markets they might not be able to justify (to themselves, at least) the significant expense of a competitive 64 bit design. Then again, maybe with today's techniques it wouldn't be so hard to simply widen the Cortex A9, which is already up to 4-way SMP and out-of-order superscalar. Achieving those two features wasn't cheap, I'm sure.
# Synthesizable high-performance 64 bit RISC architecture (ARMv6)
# 64 bit ARM® Instruction Set
http://www.toshiba-components.com/ASIC/ARM1176JZF-S.htm
though thats pretty much the only mention a quick search coughs up, so I wouldn't call it definitive, for one thing I don't know if the 1176 is in production of not.
edit: the ARM website doesn't seem to mention 64 bit anywhere either so toshiba may be engaging in a rather creative definition of 64bit.
The 1176 has been in production for several years now, but is a 32-bit part.