ARM Servers getting ready to disrupt Intel’s $50Billion/year server market

Posted by Charbax – November 6, 2011

10x less power consumption, 40x less cables, 10x less switches, 20x less racks, 4x more servers for 3x lower cost.

HP, the biggest Server maker in the world, is launching the ARM Powered Project Moonshot to revolutionize the server market. Together with Calxeda, they are launching the new custom designed Quad-core ARM Cortex-A9 EnergyCore processor that can be stuffed in a completely redesigned server rack to offer many more servers in a much smaller space and consuming much less power at a much lower cost.

You can be sure Google, Facebook, Amazon are looking into using these instead of Intel servers as soon as possible.

Now that Intel is losing the battle to powering the client device, they are also about to loose the battle to powering the cloud.

One little warning though. HP is Intel's biggest Server customer today. Intel provides most of the server processors for HP's $16 Billion per year server business today. So you never know what kinds of threats or "incentives" Intel might come up with now that HP has announced the Project Moonshot and Intel might try to lure HP into getting a discount on current server chips and using the Intel Atom instead. Expect Google, IBM, Dell and others to soon announce their own ARM Powered server projects also.

  • Anonymous

    Uh, you misspelled “losing” as “loosing”…

    Definitely good news for ARM to get larger share of the server market, and brings some much needed competition to the market, but still a bit premature to be calling Winners and Losers.

    ARM still has to develop higher performing solutions for applications that don’t rely on parallel computing, ARM is still a few more years away from getting 64bit to market to be applicable for a wider range of uses, and ARM still has to contend with the lack of legacy support and whether MS will provide full support that will effect how easy it would be for companies to make the switch.

    All in addition to possible “incentives” and other challenges Intel may come up with…  Though Intel should definitely be worried.

  • Yup, I wonder how many % of Google’s servers could run ARM now, in the next few months and not before 64-bit etc is ready. Not knowing much about how Google programs their servers to provide search, youtube on-demand streaming, google+ real-time AJAXified stuff, Google Docs and Apps for enterprise, Google App Engine, my guess is Google can maybe find a way to have a big chunk of that run on these current ARM Cortex-A9 servers now already, or that it wouldn’t take their programmers many months to port things over so they can run those there. It’s a big deal for them if they can do it. Basically server power consumption and cooling is probably one of the biggest expenses for Google today, other than salaries, buildings and acquisitions like Motorola etc. The growth of Google sites like YouTube, Google+, Google Docs, Google App Engine, is fenomenal, and if they can run things with 10x less power consumption, using ARM in their servers would be their biggest priority right now. And I’d guess Google might be one of Intel’s biggest server customers today. So if from one day to the next, Google finds a way to replace 25%, 50% or more of their Intel powered servers with ARM, that would be very disruptive.

    My guess is maybe Google can find out a way to still use Intel for where they do need higher performance, and use ARM everywhere else, and thus within their server infrastructure, combine those automatically and get the most out of each of them.

  • Maventwo

    HP Redstone servers with CalXeda ARM processors especially when they comes in 64-bit processors they will put Intel out of server business with this 4watt processors with at least 4 cores but Cortex A15 are developed for up to 8 cores.

    HP Redstone servers will be even more interesting forthcoming years with RRAM memories like HP Labs developed Memristor or Unisemi and some other RRAM memory developed technologies.

    HP Labs Memristor uses TiO2=Titanium-Oxide in theirs knots or memory nodes, on youtube you can find some interesting lectures about RRAM memories even Stan Williams who is HP Labs manager for the development of the Memristor.

    Stan Williams present the facts of the Memristor that in writes in 2ns and HP Labs have developed Memristor in up to 8 layers but they think that it cvan be possible to develope Memristors up to 1000 layers.
    But how will it be possible to make Memristors in 1000 layers, will the wafer be in the fab for a week?!

    On eetimes dot com in their recently artik´ckle about fortcoming RRAM memories they say that Samsung have also developed RRAM memories and Samsung say that RRAM memories will be able to fix 1 Trillion write-cycles to compete with mlc0multi-layer-cell in todays Nand Flash memories that fix only 1 million write-cycles.

    Stan Williams say in one of the videos on youtube that Memristor will last for beyond 10 years.

    Memristor and other RRAM memories tech memories with the memory nodes change from amorphus state to crystalline in different levels where the current changes from amorphus or crystalline or crystalline to amorphus state (if you erase).

    In July I read on engadget dot com that IBM Research Lab in Zürich announced that the think that RRAM memories will outcompete both HDD and Nand Flash memories in 2016.

    RRAM memories will be msch more cheap in larger volymes forthcoming years because RRAM tech is much easier to make in the Fabs than Nand Flash memories.

  • Porusmehta

    Hi, I agree with CyberGusa, performance matters. Today a lot of people think about web applications as the benchmark fr the world’s servers. However, that’s not always the case. Many telecom based applications (I work in telecom as a software engineer) require higher processing capability that only Intel can provide today. There is one low power RISC technology out there that does not use ARM designs- SPARC. I know first hand that Oracle (Sun’s) SPARC processors simply can’t provide the kind of performance that Intel can. I’m extremely disappointed with the performance of RISC processors and Oracle has been losing market share to Intel + HP. So unless ARM chips can bring real processing power i.e. each core has a good clock speed above 2Ghz and do some serious crunching, I can’t see ARM challenging Intel in this segment. There’s also code compatibility. I doubt we are going to get Java, Oracle, and a whole bunch of other stuff for ARM very soon.

  • The bulk of the server growth happens in cloud services and companies like Google can probably figure out to parallel process most of the stuff they do. They’ll still have the Intel processors where those are required but to serve most other things, such as serve web pages, serve data, grid compute, the ARM servers can save Google 10x the power consumption in their data centers which is currently their biggest cost.

  • Pingback: Applied Micro X-Gene ARMv8 64-bit Server-on-chip shown on FPGA –

  • Anonymous

    Biggest problem is they aren’t nearly as fast as a regular x86 server….

  • Pingback: ARM Powered Server Calxeda EnergyCore launch video –

  • Pingback: Highlights of 2011 on –

  • Pingback: » Applied Micro’s ‘X Factor’()