Sign in with
Sign up | Sign in

Your Smartphone CPU May Go into Data Centers

By - Source: Tom's Hardware US | B 15 comments

An interesting collection of firms have invested in a new ARM licensee targeting the data center.

There are more than 100 licensees of ARM's core logic semiconductor technology. So what makes Smooth-Stone unique? For starters, it's going after a very different market than your typical ARM licensee, and second, it's got quite a collection of investors.

The Austin, Tex.-based firm announced this week it has raised $48 million from a cadre of investors that range from ARM itself to several venture capital firms, including Advanced Technology Investment Company (ATIC), the company that rescued AMD and helped it spin off Globalfoundries, plus electronics giant Texas Instruments.

"This kind of investment, the amount, and the strength of this syndicate is a strong endorsement for the innovation we are bringing to market," said Smooth-Stone CEO Barry Evans in a statement. "We look forward to taking advantage of the insights and know-how of these industry-leading investors."

The vast majority of ARM licensees target either embedded systems or handheld devices, thanks to its low-power design. Apple is the company's most famous licensee, with ARM-based processors in all of its handheld devices.

But Smooth-Stone plans to bring the power efficiency of the ARM processor to servers and data centers, traditionally the haven of Intel and AMD. Smooth-Stone said it will design semiconductors and software focused on energy efficiency and increasing compute density in the data center. Smooth-Stone said its customers will have "new, unseen options as they plan their future server deployments."

Power efficiency is the new watchword for data center managers and administrators. As data centers have grown in complexity and density, the biggest challenge has been electricity, not compute power. Admins could always cram in more hardware but in many locations, such as dense urban areas like New York City, can't get any more power to run the hardware.

"Our goal is to completely remove power consumption as an issue for the data center. Imagine that change for companies with a large presence on the Internet," said Evans. "They all deal with the reality that as the mass of information grows daily, so does their power consumption. Every day these companies are thinking about managing their data center sprawl. We want to make sure that space and power are not constraining their potential."

Smooth-Stone's idea is not new. SeaMicro, a Santa Clara, Calif. company, recently announced ultra-dense servers using Intel's Atom processor that it claims would cut power and space requirements by up to 75 percent. Boston-based Tilera, spun out from MIT, has announced its own processor that offers a similar low-power 64-core processor for the data center.

Both firms have the same basic premise: for certain server functions, like Web serving, a Xeon or Opteron processor is overkill. So they offer use low-power, basic processors to do simple page rendering in a fraction of the power draw of a Xeon or Opteron.

But taking on Intel and AMD with $48 million in VC money? Good luck, said In-Stat analyst Jim McGregor. "$48 million is enough to put a sign on the door for a month. You can't do a chip on that," he said.

"It's a hard stretch for ARM to catch Intel in the data center. I think ARM has the best potential in embedded and other home consumer electronics. It's a natural for them because of their low power architecture. Scaling up into the data center is a much bigger stretch," he added.

It's not hopeless, McGregor added. There are future ARM products that have not been announced, and the company may come out with something much more powerful than it has now. But he thinks the firm would be better served targeting embedded instead of the enterprise.

"Future products from ARM will have a much broader appeal because ARM is being pushed by partners to go after other market segments," said McGregor.

Display 15 Comments.
This thread is closed for comments
  • 0 Hide
    LLJones , August 26, 2010 11:54 PM
    Have to agree. 48 mill vs several billion and well established is an uphill climb.
  • 4 Hide
    Haserath , August 27, 2010 12:07 AM
    Quote:
    Your Smartphone CPU May Go into Data Centers

    I'm sorry, that would be called stealing :)  .

    48mil does seem low to be able to design a chip that will go against a 10 core Xeon or 16 core MagnyCour. The adoption will happen slowly even if they are able to design a good chip that could compete. They should look for a market that wouldn't be as hard to enter, just so they could get more money before they try to take on the big boys in the server market.
  • 0 Hide
    Firehead2k , August 27, 2010 12:15 AM
    The minimalistic budged aside, I'm sure there is demand for these specialized architectures.
    However I can't imagine entire data centers running only on ARM powered devices that have been engineered to do one thing extremely efficient.
    A scenario would be rendering HTML5 pages, but a year from now the big players decide to add something and make revision 5.1 that won't work with the existing hardware (or run very inefficient).

    I suppose in the end it will depend on the companies cost/benefit requirements. If for example a Xenon can render 100 pages in a set time, and the specialized hardware can do 10000 with the same power draw even if they have to replace the HW within 2 years, it might just be feasible.
  • 4 Hide
    palladin9479 , August 27, 2010 1:20 AM
    The idea here is that you can also add more ARM CPU's in less space then conventional x86 systems. Just imagine your smart-phone is housing a CPU, memory, permanent storage (usually SD flash), battery and a big nice display to show it all off. The CPU itself is ultra tiny and use's very little power with almost no head load (comparatively). This makes it ideal for massive multiprocessing. With the right architectural engineering you can throw 16 or more of these cores, hell 32 might be possibly in a year or so, on a single blade. Putting dozens and dozens of blades together and you can easily establish a mini-super computer that uses incredibly small amounts of power and can be easily reconfigured for any task. You would "upgrade" by swapping out blades or just adding more since ARM is an architecture not a specific CPU. And the best part is you can easily turn off single CPUs or entire cards if their power isn't being used or otherwise needed where as current CISC x86 CPUs struggle with that due their monolithic architecture.
  • 0 Hide
    fayzaan , August 27, 2010 1:31 AM
    Ill buy the Fitness SXXXXX for $20 pls and thank you
  • 0 Hide
    kelemvor4 , August 27, 2010 3:03 AM
    Every 5 or 10 years we hear big processor news about risc is going to really take off and replace some sector currently dominated by cisc. It hasn't happened yet. I remember risc discussions were all the rage in the 90's on FIDOnet before this whole internet thing really got going... Some things never change.

    I bet the duke nukem forever servers will all be risc based arm processors...
  • 1 Hide
    hemelskonijn , August 27, 2010 3:27 AM
    Am i the only one that loves AMD/ATI products but every time i read about Global foundries imagines it to be like Veridian Dynamic?, i don't know but Global foundries just sounds evil to me.
  • 0 Hide
    dEAne , August 27, 2010 3:35 AM
    yeah.
  • 0 Hide
    ceteras , August 27, 2010 9:56 AM
    They don't need to develop a new CPU, it's already done by ARM. All they need is a technology to allow feeding thousands of these hungry little CPU's with data.
  • 0 Hide
    eyemaster , August 27, 2010 1:30 PM
    It won't matter that they only have 50$ mil. Once the 'product' is proven, they can license it away to other companies such as AMD and Intel.
  • 1 Hide
    dgingeri , August 27, 2010 3:07 PM
    ARM chips could have a great co-existence with the current data center hardware with 3 markets:

    1. IPMI processors - this could be the first break-in point. Currently, server manufacturers use some pretty bad hardware for this purpose. With ARM chips, it could be done with an industry standard, less power, and easier maintenance.

    2. hardware acceleration for RAID arrays - current chips used in RAID arrays have less calculation speed and use far more power than ARM chips. this could be the second break-in point

    3. IP KVM management - for those systems without IPMI, management can be trouble, and many don't even like KVMs (like IBM P series machines and many HPUX machines) Having a system with constant responses for the keyboard and mouse being connected, so the system doesn't drop keyboard and mouse support, would be vital for constant support. An ARM based machine could maintain this and allow for remote IP KVM management that could be handled from any other computer. It would save a whole lot of frustration for admins.

    Sure, doing mainstream server duties would be nice for an ARM based system, but that will take a while to break in. Using these markets to break in first would be a smarter decision. Then, once the company is profitable, start pushing for high duties. This is how Intel got into the server industry, starting way back with 286 machines being used as local access terminals.
  • 0 Hide
    descendency , August 27, 2010 3:29 PM
    Even if they produce quality CPUs, it will be a long time before anyone takes them seriously.
  • 1 Hide
    Anonymous , August 27, 2010 6:59 PM
    as long as the ran program is multi threaded (optimized for many threads) ARM servers could benefit all others.
  • -1 Hide
    kamen , August 30, 2010 3:22 PM
    Instead of developing ARM based architecture to compete against Intel and AMD in the server market they should develop an all in one server appliance. As much as I can't stand anything Apple I feel that creating a low cost, low power web server appliance along the concept line of an "iserver" would be a winner. This product would come with everything needed right out of the box, just add your pre-built website and voila... push button ecommerece for the masses. Sell one for $400 bucks a piece with the ability to link them together for future growth. Who needs ebay? everybody on my block comes to me to sell their stuff.
  • 0 Hide
    palladin9479 , September 1, 2010 1:23 AM
    @Kamen no serious business buys all-in-one web appliances for their web hosting. Not if that hosting is a critical part of their business infrastructure. Most just our-source to a data center, and they most definitely don't use the all-in-one set-top-appliance method. There is a very very good reason Apple never broke into the business sector and stayed almost purely consumer.

    The device needs customization and easy reconfiguration, no two customers have the exact same prioriys / needs. Thus the solution must be tailored to the problem. Try to think of IT / Tech as creating solutions for present or future problems.