HP Building Servers Using ARM-based SoCs
HP and Calxeda have teamed up to build servers based on ARM's low-power design.
Dow Jones Newswires reports that HP has teamed up with semiconductor start-up Calxeda to develop servers based on ultra low-power ARM chips. These servers will be focused on companies who build large data centers and need to lower both their physical footprint and overall energy consumption. These companies include those who deal with cloud computing, the Internet, and those looking to do analysis on their data.
According to sources close to the project, HP and Calxeda will soon unveil a prototype server and plans for a proof-of-concept program as well as more details about partnerships. Sample chips produced by the partnership will likely make an appearance by the end of the year, and then ramp up to a full-fledged volume production by the second half of 2012. These chips will consume about 90-percent less energy, take up around 90-percent less space and have a lower overall cost of ownership compared to Intel's mid-range server processors.
ARM is actually an investor in the Austin, Texas-based Calxeda. According to reports, the first reference design will be based on an ARM Cortex-A9 quad-core SoC. Server builders will be able to design systems as dense as 120 ARM quad-core nodes (480 cores) in a 2U enclosure, with an average consumption of around 5W per node (1.25W per core). The chips may be manufactured at Globalfoundries using 45-nm or 28-nm process technology.
Naturally HP, ARM and Calxeda declined to comment on partnerships that have not been made public. However last week ARM Vice President Michael Inglis said that ARM-based chips will first appear in server machines used to support basic access to websites, and then move up to more powerful systems. "As we move forward into 2014, you'll begin to see systems emerging," he said.
Meanwhile Intel seemingly doesn't take ARM's entry into the server market as an immediate threat. "We don't take any threats to our server business lightly, but there are a number of challenges for the ARM architecture to be successful in the server market," Intel spokesman Bill Calder said. "We believe the best-performing platform will win."
In addition to working with HP, unnamed sources report that Calxeda is also talking with other major server makers, storage vendors and other companies about using its processors in their products. Partnerships are expected to be announced within the next few months. Karl Freund, Calxeda vice president of marketing, said the company is in various stages of discussions with many partners about bringing products to market.
nicely put
this could honestly be the begining of a power shift for arm and intel.
if arm succedes in the server market, it will cross to the desktop, now, 1 arm cpu wont really be powerfull enough there, i was trying to figure out how much the cpus cost, but google has been failing me there, however i did come across this, "$0.10 per processor shipped" its the cost that they pay arm on average per chip, or how much they get, so im assumeing that r&d for chips is sinificantly less of a cost than for intel and amd,
now im assumeing that you can slap 4 of these quad core ones into a single computer and come out less than a single quad form intel.
an essentially noiseless cpu with low heat? 16 threads? ass time evolves its obvious that applications WILL be threaded, once that happens, will intel/amd be able to compete with what arm has to offer? they may not be able to compete on clock per clock, but on the same power draw of a quad core intel sandy bridge, or amd phenom II
intell being 95 watt
amd being 125 watt
arm being 5 watt
you can fit 19-25 cpus in a pc at same powerdraw, space may be an issue there though, and no idea on cost,
[ SOS ] Complaint about Human Rights Violations by IBM China on Centennial
Please Google:
IBM detained mother of ex-employee on the day of centennial
or
How Much IBM Can Get Away with is the Responsibility of the Media
or
Tragedy of Labor Rights Repression in IBM China
IF this is true, then is ARM really 10x better than Intel? If so, Intel needs to do more than have better manufacturing. Can someone elaborate?
(I personally still see HP going down hill from a R&D powerhouse to a repackaging company).
thanks!
A *lot* of server processes are entirely gated by network speed, like web servers, for example. In those cases, a large number of low-power cores can help with making the system as responsive as possible under heavy load.
On the other hand, database servers are typically a lot more CPU-power-bound, so the fastest possible processor is a better fit there.
And btw, the $0.10 is the license fee ARM manufacturers pay ARM for use of their IP per core shipped.
The major difference between ARM and Intel lies in their prospected roadmaps. ARM expects at least a doubling in processor power every for at least the next 5 years in the same power envelope. Intel expect a maximum 20% rise in their processing power per year on average. So even though Intel had an enormous head start 3 years ago, they are really being leapfrogged.
when i say 10 cents, im talking about what they pay arm. as i have gone into lenghy discussions with people in the past over what makes a cpu cost so much. we came to the conclution that the bulk of the cpu cost deals with r&d, with a 10 cents r&d cost per chip on arm, i cant see ir being all to much more expensive than manufacture cost lets say it costs 50000 to manufacture the chips, and you have a yeild of about 1000, that would be about 50.10$ to make. proffit would come in at about 10$ a chip markup, and a further 10$ by whoever buys it to sell, or put in a device. all in all 70$
im assumeing all this of coarse, because the yeild in a batch may cost less to produce, and may have higher yeilds, as based the 50k off a figure i heard on i think a amd fabrication.
"Clock for Clock"...are we talking integer operations? floating point operations? AVX?? You can't just say 1 processor is "faster" than another "clock for clock" without telling exactly what type of instruction/operation is used. For ARM to actually compete against Intel in the desktop space, they will have to start adding support for multimedia instruction sets.
How many ARM chips will it take to equal the performance of a 95 Watt Intel 2700K 3.9Ghz Quad Core? How many glue logic/chips will you need to link dozens of ARM chips? There are so many features ARM needs to add to compete in the server market that it won't be small and cool chips anymore.
Intel 2700K (Quad core) @ 3.4Ghz - 128,300 MIPS
ARM Cortex-A15 (Quad core) @ 2.5GHz - 35,000 MIPS
ARM Cortex A9 (Dual core) @ 1.0GHz - 5,000 MIPS
MIPS isn't very accurate but it's one metric.
Cortex-A15 isn't even out yet and you're looking at 4 of them to equal a $300 Intel CPU.
I've seen a 16 X Cortex-A9 @600Mhz server that uses only 80 Watts but that's only 1/3 of the compute power combined.
http://www.linuxfordevices.com/c/a/News/ZT-Systems-R1801e-/
By the time ARM adds 64bit capability it will be competing with 14nm Haswell.
So as a engineer I have to choose the best platform for whatever workload I'm engineering for. When you need an incredible amount of distributed parallel operations at once, yet the actual execution speed isn't a big concern (wide and shallow performance footprint) then something like ARM makes sense. If you need a smaller amount of operations but those operations execution speed is time sensitive (narrow and deep), then a x86 platform makes sense. And if you need a large amount of parallel operations that are connected and time sensitive (wide and deep) then something like SPARC is a better option.
It's impossible to state that one architecture is better then another without including the context of the comparison.