I like the concept, but something has to change for this to take off. There are already low-power Xeons at at least one high-density many-node Atom solution (Sea Micro's SM10000) but they haven't taken over.
One difficulty for ARM is that servers have lots of other power-hungry and pricey components so a CPU can't radically boost efficiency on its own. And instruction sets matter, though anything Linux is pretty portable and Windows' kernel has been ported. And Intel ain't giving ground here without a fight.
In favor of ARM: lots of work is being done on ARM speed. Datacenter workloads adapt better than most to many cores or many servers. Energy prices could spike in some places/at some times -- picture a CO2 tax, a move away from nuclear power, geopolitical trouble getting fuel, or a combination of those. And not only electrical costs but density favors low-power servers. If ARM chips sell at lower profit margins than x86 ones, that could be key, too.
I suspect there's plenty of "fat" in the datacenter -- servers that are at low CPU utilization and always will be because they're I/O or RAM-bound. Those might be candidates for ARM chips. Finally, ARM doesn't actually have to take over the DC to be important. If it starts making inroads in particular niches or for a small percent of the market, Intel's gonna feel threatened and do something exciting in response. :)
Currently using a dockstar for my file-serving and torrenting needs. Something with a bit more punch behind it in a low power envelope would be highly welcome.
You write that: >Of course, performance is a big question mark but if AppliedMicro's tests are to [be believed] (little typo), X-Gene is up to three times faster than Intel's Sandy Bridge based E3 Xeons when looking at similar power profile.
That's pretty meaningless as written though. Even on a server, not everything can be parallelized to an unlimited degree. Per core performance can still matter, even without touching how well it works for int vs fp vs highly vectorized loads. This sounds like an exciting announcement, but I'm inclined to take these marketing claims with a grain or ten of salt until hard tests start popping up. It will no doubt push forward what can be done by ARM though, and any new sources of competition are always great to see.
All of these numbers should definitely be taken with grain of salt. Like I mentioned, they are all based on pre-silicon projections (and given by APM, so no unbiased source) and it will be a year before we see the first samples.
How long before one of them acquires an ARM licence?
If AMD decides to adopt ARM, think of the number of AMD-badged ARM solutions that would be produced each month through GloFo and TSMC. What's more, with an architecture far closer to what we're seeing in most portable solutions, a shift to ARM might even create a new renaissance of easily ported software. Also, if GloFo goes into major production of ARM chips, it'd take far less time to produce chips as well as locate and fix problems than with the comparatively large x86/64 dies we're seeing. Intel would have to at least consider changing its plans, but neither Intel nor AMD would relish at the prospect of a huge amount of wasted money on their current and future plans and a switch to something else. Change can be very slow if change happens at all, and this would be a complete paradigm shift if x86 was thrown away after 33 years.
Of course, I don't know what I'm talking about, so feel free to slap me aside the face if you disagree. :P
I don't hate x86, but we can't keep using it forever. Perhaps Intel already has an ARM-esque architecture planned.
For both AMD and Intel it would seem wise to have an ARM license on the side. My guess would be they are discussing this over lunch. Then again, there's such a wide and deep legacy in x86. Abandoning x86 will take longer than abandoning PATA or PCI and even those are not quite gone. I think both the ARM hardware and softwarebase has to mature to a certain point where manufacturers and customers alike can decide if there's still a future for x86 or if the two can happily coexist. The end of it is nowhere near.
Intel has ARM licenses, Intel's wireless business (Ex Infineon maker of the popular X-gold baseband) still has licenses for new cores as well as old. Also app-processors. As well as all the licenses for their raid-processors that is ARM. ARM sold their own lineup off Application Processors to Marvell though. They are no newcomers.
AMD had made products for ARM-environment like the GPUs driving Qualcomms chips and Broadcoms graphics processors, but they sold that off. Their complete chip (ATi) solutions was for MIPS though and AMD themselves has previously made several MIPS-variants that it has divested. Neither of the companies are strangers to SoCs, multimedia SoCs, Network SoCs and other embedded SoCs. But they would have no benefit from competing with the companies they sold their mobile IP off to.
So don't treat them like x86 is all they know and have done. Both invested heavily in RISC in the 80's, Intels i960 and i860 never took off as a general processor. Intel Xscale was vastly popular in smartphones, consumer products and industrial products though. AMD began it's processor design with AMD 29K and they were used in a few mainframe and high-end/minicomputer projects as well as embedded designs and as well as was their basis for a 586/Pentium compatible processor. It's just totally history-less and clueless.
Intel has little interest in ready made cores or their own ARM-designs today. They can just as well do x86 for SoC and embedded market. There is also lot of other things that needs to work in order to have a strong CE SoC. AMD does seem to have exited the low-power embedded market all together if you count out their AMD64-products. But they do seem to target consumer products. Porting is already done and financially supported by companies TI, Freescale, ST-E, STM, NXP, SHARP, Marvell, Samsung, Qualcomm, Broadcom, Atmel, Fujitsu, Renasas and others and Software companies like Intel through Wind River, IBM etc.
Any word of what type of memory these chips will support? A server with 128 cores is going to need a lot of memory. I'm also curious to see how well these ARM chips run virtualized workloads.
This is an exciting time in the world of processors. After a few years of relative stagnation, competition is finally heating up. My the best chip(s) win!
They'll have the monopoly for the base licenses which are very cheap. But there are a dozen or so ARM chip makers, so from a competition point of view, it's a whole lot better than the x86 market.
Intel will get there by the time ARM make first prototype, considering their core is many times more powerful than ARM. This is not good enough reason for people to switch.
Also don't forget about the uncore and memory power will take up over half total system power. It will further dilute the core power advantage. Idle power is supposed to be good with Tri-gate. And Intel can employ the trick they used in mobile chip, just turn off part of the silicon.
As ARM is pushing performance, they also suck more power. They don't have secret sauce.
A core many times more powerful than ARM's won't do much good. Just like the ARM, it will eat power with the uncore. For a given amount of work, both will be limited in performance by cache associativity and size, and main memory.
ARM may not have a secret sauce, but neither does Intel, in this situation. These kinds of systems will be squarely aimed at market niches that want efficient distributed I/O, far more than they want CPU power. ARM and ARM vendors do have one advantage, however: being used to lower margins than Intel.
An unsecured business loan can be an excellent tool for expanding your company and making more interest in your product or service. It has many perks over a secured type of loan such as no requirement for you to put up assets as collateral. === http://www.fundfactor.com ===
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
22 Comments
Back to Article
DesktopMan - Friday, October 28, 2011 - link
And so it begins...twotwotwo - Friday, October 28, 2011 - link
I like the concept, but something has to change for this to take off. There are already low-power Xeons at at least one high-density many-node Atom solution (Sea Micro's SM10000) but they haven't taken over.One difficulty for ARM is that servers have lots of other power-hungry and pricey components so a CPU can't radically boost efficiency on its own. And instruction sets matter, though anything Linux is pretty portable and Windows' kernel has been ported. And Intel ain't giving ground here without a fight.
In favor of ARM: lots of work is being done on ARM speed. Datacenter workloads adapt better than most to many cores or many servers. Energy prices could spike in some places/at some times -- picture a CO2 tax, a move away from nuclear power, geopolitical trouble getting fuel, or a combination of those. And not only electrical costs but density favors low-power servers. If ARM chips sell at lower profit margins than x86 ones, that could be key, too.
I suspect there's plenty of "fat" in the datacenter -- servers that are at low CPU utilization and always will be because they're I/O or RAM-bound. Those might be candidates for ARM chips. Finally, ARM doesn't actually have to take over the DC to be important. If it starts making inroads in particular niches or for a small percent of the market, Intel's gonna feel threatened and do something exciting in response. :)
gevorg - Friday, October 28, 2011 - link
This is why its hard for Intel to compete with ARM:"Up to 2W per core"
Polymerabbit - Saturday, October 29, 2011 - link
Mobile Broadwell processors should be able to achieve that. Keep in mind that this stuff is set to be released in 2014.silverblue - Saturday, October 29, 2011 - link
Yes, but 0.5W idle? I know there's a lot of cores, but can't they be shut off entirely to conserve power even further?Soulkeeper - Friday, October 28, 2011 - link
16 or 32 cores targeted at me would be good :)iwod - Friday, October 28, 2011 - link
How about just give me 2 Core + 1G Ethernet and a Raid Controller for NAS.With ARMv8 this should be 4 to 5 times faster then today's Marvell Kirkwood NAS CPU.
mlcloud - Friday, October 28, 2011 - link
Currently using a dockstar for my file-serving and torrenting needs. Something with a bit more punch behind it in a low power envelope would be highly welcome.zanon - Saturday, October 29, 2011 - link
You write that:>Of course, performance is a big question mark but if AppliedMicro's tests are to [be believed] (little typo), X-Gene is up to three times faster than Intel's Sandy Bridge based E3 Xeons when looking at similar power profile.
That's pretty meaningless as written though. Even on a server, not everything can be parallelized to an unlimited degree. Per core performance can still matter, even without touching how well it works for int vs fp vs highly vectorized loads. This sounds like an exciting announcement, but I'm inclined to take these marketing claims with a grain or ten of salt until hard tests start popping up. It will no doubt push forward what can be done by ARM though, and any new sources of competition are always great to see.
Kristian Vättö - Saturday, October 29, 2011 - link
All of these numbers should definitely be taken with grain of salt. Like I mentioned, they are all based on pre-silicon projections (and given by APM, so no unbiased source) and it will be a year before we see the first samples.silverblue - Saturday, October 29, 2011 - link
How long before one of them acquires an ARM licence?If AMD decides to adopt ARM, think of the number of AMD-badged ARM solutions that would be produced each month through GloFo and TSMC. What's more, with an architecture far closer to what we're seeing in most portable solutions, a shift to ARM might even create a new renaissance of easily ported software. Also, if GloFo goes into major production of ARM chips, it'd take far less time to produce chips as well as locate and fix problems than with the comparatively large x86/64 dies we're seeing. Intel would have to at least consider changing its plans, but neither Intel nor AMD would relish at the prospect of a huge amount of wasted money on their current and future plans and a switch to something else. Change can be very slow if change happens at all, and this would be a complete paradigm shift if x86 was thrown away after 33 years.
Of course, I don't know what I'm talking about, so feel free to slap me aside the face if you disagree. :P
I don't hate x86, but we can't keep using it forever. Perhaps Intel already has an ARM-esque architecture planned.
cfaalm - Saturday, October 29, 2011 - link
For both AMD and Intel it would seem wise to have an ARM license on the side. My guess would be they are discussing this over lunch. Then again, there's such a wide and deep legacy in x86. Abandoning x86 will take longer than abandoning PATA or PCI and even those are not quite gone. I think both the ARM hardware and softwarebase has to mature to a certain point where manufacturers and customers alike can decide if there's still a future for x86 or if the two can happily coexist. The end of it is nowhere near.Penti - Saturday, October 29, 2011 - link
Intel has ARM licenses, Intel's wireless business (Ex Infineon maker of the popular X-gold baseband) still has licenses for new cores as well as old. Also app-processors. As well as all the licenses for their raid-processors that is ARM. ARM sold their own lineup off Application Processors to Marvell though. They are no newcomers.AMD had made products for ARM-environment like the GPUs driving Qualcomms chips and Broadcoms graphics processors, but they sold that off. Their complete chip (ATi) solutions was for MIPS though and AMD themselves has previously made several MIPS-variants that it has divested. Neither of the companies are strangers to SoCs, multimedia SoCs, Network SoCs and other embedded SoCs. But they would have no benefit from competing with the companies they sold their mobile IP off to.
So don't treat them like x86 is all they know and have done. Both invested heavily in RISC in the 80's, Intels i960 and i860 never took off as a general processor. Intel Xscale was vastly popular in smartphones, consumer products and industrial products though. AMD began it's processor design with AMD 29K and they were used in a few mainframe and high-end/minicomputer projects as well as embedded designs and as well as was their basis for a 586/Pentium compatible processor. It's just totally history-less and clueless.
Intel has little interest in ready made cores or their own ARM-designs today. They can just as well do x86 for SoC and embedded market. There is also lot of other things that needs to work in order to have a strong CE SoC. AMD does seem to have exited the low-power embedded market all together if you count out their AMD64-products. But they do seem to target consumer products. Porting is already done and financially supported by companies TI, Freescale, ST-E, STM, NXP, SHARP, Marvell, Samsung, Qualcomm, Broadcom, Atmel, Fujitsu, Renasas and others and Software companies like Intel through Wind River, IBM etc.
dcollins - Saturday, October 29, 2011 - link
Any word of what type of memory these chips will support? A server with 128 cores is going to need a lot of memory. I'm also curious to see how well these ARM chips run virtualized workloads.This is an exciting time in the world of processors. After a few years of relative stagnation, competition is finally heating up. My the best chip(s) win!
nofumble62 - Sunday, October 30, 2011 - link
ARM will be the new monopoly and the licensing fee will increase to everybody.Like a dog chasing its tail once again, hehe
Lucian Armasu - Friday, November 4, 2011 - link
They'll have the monopoly for the base licenses which are very cheap. But there are a dozen or so ARM chip makers, so from a competition point of view, it's a whole lot better than the x86 market.zhangqq - Monday, October 31, 2011 - link
http://ygn.me/bTf7pnofumble62 - Tuesday, November 1, 2011 - link
Intel will get there by the time ARM make first prototype, considering their core is many times more powerful than ARM. This is not good enough reason for people to switch.Also don't forget about the uncore and memory power will take up over half total system power. It will further dilute the core power advantage. Idle power is supposed to be good with Tri-gate. And Intel can employ the trick they used in mobile chip, just turn off part of the silicon.
As ARM is pushing performance, they also suck more power. They don't have secret sauce.
Cerb - Thursday, November 3, 2011 - link
A core many times more powerful than ARM's won't do much good. Just like the ARM, it will eat power with the uncore. For a given amount of work, both will be limited in performance by cache associativity and size, and main memory.ARM may not have a secret sauce, but neither does Intel, in this situation. These kinds of systems will be squarely aimed at market niches that want efficient distributed I/O, far more than they want CPU power. ARM and ARM vendors do have one advantage, however: being used to lower margins than Intel.
Lucian Armasu - Friday, November 4, 2011 - link
According to this test here, at the same performance level, a 12-core Cortex A9 chip uses 5x less power than an Intel Quad Core Xeon chip:http://www.cnx-software.com/2011/11/04/pandaboard-...
lucypinderol - Tuesday, November 1, 2011 - link
An unsecured business loan can be an excellent tool for expanding your company and making more interest in your product or service. It has many perks over a secured type of loan such as no requirement for you to put up assets as collateral. === http://www.fundfactor.com ===Arnulf - Tuesday, November 1, 2011 - link
Stupid spammers ... If I needed a loan I most certainly wouldn't consider online scammers sych as yourselves as a potential source for it.