News & Analysis
Intel microserver leaves door open for ARM
Rick Merritt
2/7/2013 3:01 AM EST
SANTA CLARA, Calif.--Intel’s first microserver processor is less power efficient than its existing Xeon chips, leaving a significant opportunity for alternative SoCs, according to Linley Gwennap, principal of the Linley Group.
Intel released last year the Atom S1000, also known as Centerton, a dual core chip meant to fend off mainly ARM-based server SoCs from a growing group of vendors. While it reduced power consumption to 6.3W, it does not support Ethernet, Serial ATA or USB controllers or multithreading.
“According to data Intel provided, this chip is less power efficient than its Xeon, so it seems like we are going in the wrong direction,” Gwennap said. “It’s not really a system on a chip yet, it has significantly lower performance and only uses 32-nm process technology,” he said at the Linley Data Center Conference here.
Gwennap characterized the chip as a placeholder for Avoton, a 22-nm CPU with a new Atom core. “They haven’t announced what it is yet, and it will not be in production until the second half of the year,” he added.
Linley Group recently released a report that projects alternative server SoCs will compete for an available market of $2.5 billion by 2016. That’s about 30 percent of a server processor market that the report estimated is edging toward $10 billion a year. The forecast assumes Microsoft will not have a version of Windows Server for ARM during that period.
Both sides have their challenges, Gwennap noted. ARM server SoCs need to port x86 server software, they won’t support 64-bit addressing until later this year and they have a lower single-thread performance than the x86, he said. In addition, Intel has a process edge and will use it to roll 14-nm chips in 2014.
On the other hand, the x86 architecture was designed for best single-thread performance. As such it is encumbered by complexities such as an ability to manage up to 168 instructions simultaneously in flight in the latest chips and up to 192 in the next-generation Haswell.
“There’s a lot of logic just moving data around without doing useful computations,” Gwennap said.
Click on image to enlarge.
Showing the integration gap, Gwennap compared an existing ARM server SoC to an Intel Xeon platform. A slide on that comparison follows along with slides from Calxeda and Applied Micro Circuits Corp. who participated in a panel.
Related stories:
Intel released last year the Atom S1000, also known as Centerton, a dual core chip meant to fend off mainly ARM-based server SoCs from a growing group of vendors. While it reduced power consumption to 6.3W, it does not support Ethernet, Serial ATA or USB controllers or multithreading.
“According to data Intel provided, this chip is less power efficient than its Xeon, so it seems like we are going in the wrong direction,” Gwennap said. “It’s not really a system on a chip yet, it has significantly lower performance and only uses 32-nm process technology,” he said at the Linley Data Center Conference here.
Gwennap characterized the chip as a placeholder for Avoton, a 22-nm CPU with a new Atom core. “They haven’t announced what it is yet, and it will not be in production until the second half of the year,” he added.
Linley Group recently released a report that projects alternative server SoCs will compete for an available market of $2.5 billion by 2016. That’s about 30 percent of a server processor market that the report estimated is edging toward $10 billion a year. The forecast assumes Microsoft will not have a version of Windows Server for ARM during that period.
Both sides have their challenges, Gwennap noted. ARM server SoCs need to port x86 server software, they won’t support 64-bit addressing until later this year and they have a lower single-thread performance than the x86, he said. In addition, Intel has a process edge and will use it to roll 14-nm chips in 2014.
On the other hand, the x86 architecture was designed for best single-thread performance. As such it is encumbered by complexities such as an ability to manage up to 168 instructions simultaneously in flight in the latest chips and up to 192 in the next-generation Haswell.
“There’s a lot of logic just moving data around without doing useful computations,” Gwennap said.
Click on image to enlarge.
The typical Xeon has blocks geared to support dozens of instructions in flight, said Gwennap.
Showing the integration gap, Gwennap compared an existing ARM server SoC to an Intel Xeon platform. A slide on that comparison follows along with slides from Calxeda and Applied Micro Circuits Corp. who participated in a panel.
Related stories:
|
|
|
|
|
Navigate to related information
ChipConnoisseur
2/7/2013 7:05 AM EST
Intel won't even be able to keep the server market together, let alone enter the mobile market. It doesn't look like Intel's chips be competitive with all the different ARM SoC's coming out next year (and in quad core or higher versions).
Sign in to Reply
rick.merritt
2/7/2013 11:11 AM EST
ASP disruption ahead:
“Intel is getting away with huge margins because there’s not a lot of competition, but I think we will see with more competition the prices come down as well as the power.”
--Linley Gwennap on the server CPU business
Sign in to Reply
US Made
2/7/2013 12:28 PM EST
well defined race Performance/Watt/$.
even if Intel leads, margins are going to come down...
Do not underestimate if Samsung playing in this market later....even though they are consumer electronics...deep pocket can do many things...
Sign in to Reply
de_la_rosa
2/7/2013 4:33 PM EST
ARM have no innovation what so ever. I hate them.
Sign in to Reply
jackOfManyTrades
2/8/2013 11:33 AM EST
Yes - they're only in 95% of the world's mobile phones - annual sales a measily 1.6 billion. No innovation whatsoever - which is why Intel find it so easy to compete in that market.
Sign in to Reply
iniewski
2/7/2013 5:26 PM EST
ARM has no innovation???
Sign in to Reply
help.fulguy
2/8/2013 11:18 AM EST
Rick, no surprise with you. Keep Brown nosing ARM.
Sign in to Reply
chipmonk
2/8/2013 2:19 PM EST
EE Times is now owned by UBM, a UK based Co. as is ARM.
Sign in to Reply
Bruzzer
2/8/2013 2:35 PM EST
Former Cyrix, NexGen, ARM, AMD, IDT, as discovery TA in FTC case v intel, as an analyst following the intersection of x86 and ARM I’m for ARM server. With that track record in Intel town who wouldn’t be for ARM server silicon and systems underdogs.
There is surely a play here where this analyst has stated ARM server as a viable business. Question is where are the product and price voids in Intel Xeon structure. Unlike other analyst’s who position ARM purely against Atom, this analyst believes ARM architecture can address the entire Xeon performance spread from E3 to 46xx. Dense ARM on blades is a viable high margin business. Multiple ARMs on blade is where the producer values is in ARM server; in multiple processor blades v Xeon.
In this article Linley speculates Centerton a place holder for Avoton where this analyst finds Centerton a monopolists attempt to position inferior dual core v ARM quad ready for the dumping as Avoton up to eight cores and integrated hub is launched speculated at 17w power. Where Avoton may already be launched shown by Intel on dual chip micro blades at Open Compute.
Centerton is not a place holder but presents a financial barrier verse all ARM SOCs. That is meant by Intel too suck ARM financial value out of production and distribution channels as Centerton is dumped at price less than Intel average fixed cost per unit of production. Likely in sales package with higher margin product where Centerton is essentially priced for free. So here we have competition Intel style no matter the product category.
Mike Bruzzone. Camp Marketing
Sign in to Reply
FriedrichE
2/8/2013 3:40 PM EST
Nope, thats failed the Turing test.
Sign in to Reply
Bruzzer
2/8/2013 2:54 PM EST
Linley places ARM Server at $2.5 billion by 2016 and in earlier report $3 bil.
This analyst on Sandy displacement puts ARM at perf ratio 4 1.4 GHz Quad to 1 Xeon 2 GHz Hexa at 56,917,517 units. Analyst suggests ARM value higher then Linely’s $50 placing in Atom land that cannot be sustainable outcome.
Report shows ARM server incorporates more system blocks that equates to value in price for functionality. If not ARM server may not be a viable high margin business as margin is the driver to attract foundries to rip chunks of revenue from the monopolist which is the primary attraction.
Currently 25 million Atom are produced every cycle, average price $60 value $1.5 billion. Sandy Xeon v ARM on 1:1 unit displacement revenue priced under $250 + Atom = 39,932,527 units value $4,082,044,422. Increase value of ARM blade verse Xeon to $625 including Atom on 1:1 displacement equals 50 million units value $11 billion. For ARM server to displace one quarter of this revenue delivers Linley’s revenue estimate.
Interesting quandary is ratio of ARM v Xeon to achieve similar perf; here 1:1 & in above example 4:1. Recall producer margin is in the blades not the individual components regardless of functionality.
Where one question beyond Samsung is what will attract the foundry; 50 million v 200,000,000 ARM server chips per year; and at what price & margin.
Needless to say speculation until whole platform delivered meaning optimized software. Where if this year total ARM server chips reach 50,000 units or minimally 1,000 validation systems that is a step in the right direction for achieving ARM server success.
The missing component then is you, the engineer and applications programmer, call APM, Calxeda and Boston, Marvell and Cogent and order your development system today. If you need a fast smart switch that can search and store at power well utilized ARM Server is worth developing into other than an Intel monopoly future.
Mike Bruzzone, Camp Marketing
Sign in to Reply
FriedrichE
2/8/2013 3:38 PM EST
Can anyone translate this into English?
Sign in to Reply
Pierr
2/8/2013 5:32 PM EST
I am perplexed by Linley Gwinnap's comments. I assume he is referring to Intel's Atom S1200, which was described in a release and media event in early December. Yet his characterization of the device is at odds both with what is in the release and what was presented by Intel and HP at the event. Moreover, Intel's device has already been included in HP's Gemini systems and HP has affirmed its superior power efficiency in large Scale-out computing applications. HP has communicated that ARM-based processors will be included in the project once they are available with 64b capabilities. But the ARM ecosystem is lagging here as this will not happen until later this year, from what I have read. AMCC claims a lead here, but they are so promotional that (for me) every statement requires corroboration. So is there even a Micro-server SOC market open to ARM this year of any size?
Sign in to Reply
rick.merritt
2/11/2013 11:52 AM EST
Yes, there is a realtively small 32-b it ARM server market emerging this year with systems from a handful of little known players such as Boston Ltd. (Calxeda) and Mitac and Wiwynn (Marvell) and at least one large user TK soon.
But the big juice comes with 64 bit products in 2014.
Sign in to Reply
giuann
2/8/2013 11:06 PM EST
ARM does not miss any opportunity to get free advertisement.
ARM are SOC therefore I have to develop my own chip or be to the mercy of one of the vendors.
Intel chip is a proven standard.
Sign in to Reply
Bruzzer
2/8/2013 11:30 PM EST
FredrichE;
I respond to business management inquiries.
What ARM server really means today, if you're a physicist looking for specific incidents occurring within the real time Big Data of an accelerator ARM server can do it well utilized at a fraction of the power.
If you're into analytic's ARM server with stock FPU and certainly adding a DSP accelerator, can do what FPGA acceleration does for lower power. Some believe on their development investment at a lower hardware cost, applications programming time and power.
If you're serving from cold storage ARM server offers low power for a job done well done.
If you're serving anything Web 2.0, there is from storage search and serve that ARM does well for lower power.
Penguin Computing has put up ARM pages if you look through gets down to the brass tacks about ARM server.
http://www.penguincomputing.com/search/node?keys=ARM+SERVER
Mike Bruzzone, Camp Marketing
Sign in to Reply
wsw1982
2/11/2013 2:55 AM EST
What does the author mean by power efficiency? If the author means performance/watt, the ARM, too, is not as power efficiency as Xeon.
http://www.theregister.co.uk/2012/08/13/xeon_vs_calxeda_arm_apache_bench/
Both ARM/ATOM is good at absolutely power consumption instead of power efficiency.
Sign in to Reply
cheapMonk
2/11/2013 8:28 AM EST
Please, where can i buy TODAY a microserver with an ARM inside ? (64 bit please)
Sign in to Reply
rick.merritt
2/11/2013 11:55 AM EST
Nothing exists for at least six-nine months except FPGA-based development prototypes
Sign in to Reply
ChipConnoisseur
2/11/2013 3:14 PM EST
And where can you buy a 64 bit Atom server? Nowhere. In fact, Intel will be at least a year later to this market, as they won't have a real Atom server chip (64 bit) until 2015.
Not to mention that even for that generation their transistors are really 26nm, not 22nm.
Sign in to Reply
cheapMonk
2/13/2013 3:25 AM EST
so, it's a micro...market ?
Sign in to Reply
Pete.Kuan
2/12/2013 8:34 AM EST
ARM is an open mind company, they are willing to provide all knowledge to any customers. Compared to ARM, Intel is a miser.They used all kinds of political issues to prevent others to share their apple. In the PC era stage, they successfully kick out all competitors. But its drawback is obviously showing up at mobile phone time. This is the bittermelon Intel must pay now.
Sign in to Reply
kjdsfkjdshfkdshfvc
2/12/2013 11:07 AM EST
Intel is gonna nail it.
http://bit.ly/dI3hcF
Sign in to Reply