Seeking Alpha

ARM rises, Intel slips, AppliedMicro dives following Google CPU report

  • With ARM-based (ARMH +2.6%) CPUs currently having a minimal presence in the massive server CPU market, a report that Google is thinking of developing its own ARM-based server CPUs has gone over well with ARM investors.
  • In addition to Google, Facebook has shown an interest in ARM. A recent post from a Facebook developer suggests the company is working to port some of its internal software for use with ARM CPUs.
  • AppliedMicro (AMCC -4.6%), an early leader in the ARM server CPU space (courtesy of its X-Gene chips), wound up selling off after opening higher. There may be concerns Google will open-source its CPU designs to help foster a broader ecosystem for them, as it has done with many pieces of software. AMD, Marvell, and private Calxeda are also targeting the ARM server CPU market.
  • Intel (INTC -0.7%), whose server CPU division had a $1.39B Q3 op. profit on the back of 12% Y/Y rev. growth, finished down modestly. Google/Facebook would act as big reference wins for ARM, which still face major software ecosystem challenges as it tries to grow its server presence. Intel is trying to counter ARM with its low-power Avoton Atom server CPUs.
  • FBR, which just started coverage on ARM with an Outperform, thinks ARM-based designs could grab 10% of the server CPU market by 2018, thanks to a 50% share in the fast-growing microserver segment.
Comments (11)
  • Justin Jaynes
    , contributor
    Comments (3021) | Send Message
    $FB looking at ARM
    13 Dec 2013, 06:35 PM Reply Like
  • toonies
    , contributor
    Comments (393) | Send Message
    @justin: do you remember Feldman's notice about PHP?
    13 Dec 2013, 06:53 PM Reply Like
  • Justin Jaynes
    , contributor
    Comments (3021) | Send Message


    “They’re writing in PHP. They absolutely don’t give a shit about special instructions or optimizing. They’re writing at a layer of abstraction so high, that they don’t even know what the CPU’s doing.”


    I do - but I want to see some more concrete proof before I start running down this path
    13 Dec 2013, 07:34 PM Reply Like
  • Mike Bruzzone
    , contributor
    Comments (127) | Send Message
    Face Book CTO Frank Frankovsky is on the board of directors of Calxeda. Face Book, for the most part, is a Xeon E5 2660 shop.


    Face Book's ARM Server position has been "don't bet your business on us".


    Face Book has been open to ARM beta validation hardware where cool storage makes since for 32 bit ARM v Intel anything.


    APM has 32 nm XGene samples running on FB Open Processor Hub. Yet on this open prototyping platform v Intel, APM needs to show up with the low power advantage of XGene at 28 nm.


    Intel has been very aggressive promoting open prototype vehicles at two process nodes ahead. Which has been Intel's point strategy through ARM Crush Campaign, beginning in 2010, supplemented by a cascade of surplus dumping.


    Sandy Bridge Generation Product Dumping; 76,744,510 units. Industrial financial displacement on economics; $5,219,432,098


    Current Ivy Bridge through Haswell Product Dumping 35,271,649 units. Industrial financial displacement on economics; $1,432,313,390


    Intel Inside sales rewards for exclusive dealing; Sherman Act Section 2 Intent, on the Intel financial; $3,641,000,000


    ARM 64 LAMP Stack is just barely and still into 2015. Including JSON 64 pushed into this time frame, by Oracle, unless industry pressure following this set back, at Server Summit 2013, caused the schedule to be pulled back in.


    JSON 64 is important for ARM to run Hadoop. Where volume opportunity for ARM 64 is not necessarily low power server, but low power batch processing.


    Cloud advantage #1 for ARM 64 silicon design producers is a defined operating system and application environment. Which at FB runs CentOS; still lacking ARM port last quarter . . . but expected. Where Ubunto and Fedora are moving along.


    Cloud advantage #2 for ARM 64 silicon design producers is the ability to purchase silicon in large volume, as ARM server lacks the production economic scale to effectively compete v Intel in the commercial market lacking any volume application.


    I am a fan of ARM server, especially attached to a standard reference platform that can drive silicon economics first and foremost. Lacking silicon production economics ARM server remains a boutique development.


    Mike Bruzzone
    Camp Marketing
    13 Dec 2013, 08:18 PM Reply Like
  • DoowopDave
    , contributor
    Comments (204) | Send Message
    Intel is like the surfer who's been waiting for that big wave. If they'd caught it, they'd be in for a great ride... but they missed it. Now, they're swimming madly, trying to catch up to it as it pulls away. Much easier and more fun to ride it than chase it, perhaps to no avail.
    13 Dec 2013, 09:21 PM Reply Like
  • Ajayyy
    , contributor
    Comments (310) | Send Message
    I think the difference is that, INTC is just going the route of indoors waterpark with plenty of spectacular waves that they can have a good grasp on. The befit is that even if the environment gets bad or the wind dies down, INTC will still be able to do what they do, while others will crash in the rocks. The obvious downside is the wait it takes for the waterpark to open up to business. I'm not counting out INTC.
    14 Dec 2013, 12:58 PM Reply Like
  • Mike Bruzzone
    , contributor
    Comments (127) | Send Message
    And on Google, I am for all ARM Server silicon licensees including if APM got that deal?


    However, if ARM itself is intent continuing to over subscribe the product category, that will continue too be an issue.


    Design licensees deserve Google, Face Book, and other cloud revenue on their RISK development expense.


    Which is not ARM Inc. selling another license to see what sticks. ARM Ltd. it is way too late for that.


    And Simon, if any new licensee introduces a standard interface design, or ARM figures out why a standard reference design is must verse Intel, you Simon, Ehab and Ian will meet and confer and not on my request..


    Mike Bruzzone
    Camp Marketing
    13 Dec 2013, 09:37 PM Reply Like
  • Mike Bruzzone
    , contributor
    Comments (127) | Send Message
    And Justin,


    ARM Server software first and foremost in system integration (think Novell Network VAR except now for Data Center) is a must to optimize below the abstraction layer.


    Hadoop, Ubunto, Fedora are all raw Intel ports.


    Learn the ARM hardware application code optimizations.


    More than hardware for the 'software first and foremost" systems integrator / consultant, the money too be made in ARM Server is owning for reuse the application optimization intellectual property.


    Application IP, that learning curve from optimizing applications for customer's specific use before others.


    Like Nvidia CUDA beta validation knowing all the undocumented calls on the validation.


    Except in the ARM World that opportunity has not been fixed. The leading opportunity is an open green field.


    Owning that IP is the first mover advantage for whole systems.


    Mike Bruzzone
    Camp Marketing.
    13 Dec 2013, 10:00 PM Reply Like
  • Justin Jaynes
    , contributor
    Comments (3021) | Send Message
    @Mike - thanks much for those - I always enjoy your posts and wish I could see more of them. You have some eye openers.
    13 Dec 2013, 10:52 PM Reply Like
  • Mike Bruzzone
    , contributor
    Comments (127) | Send Message
    Another way to think about the Register Post;



    Is that FB has adopted Elbrus or similar ARM x86 binary translation.


    There are actually four or five sources for the IP which for visual reference in lab is a hardware accelerated decoder card, that transforms x86 CISC instructions into the native architecture's RISC operands.


    That IP block is then hard implemented into the processor.
    P6 is an early example.


    Binary translation in its software state is a form of emulation. Emulation that allows Alpha and later NexGen 586, AMD K6 (Nx686), Cyrix M1 (I think), IDT C6 and now ARM to perform some x86 instructions that does add to processor overhead.


    For Cyrix 486 (GM:CISC) and Nx586 (RISC) only the most common Windows instructions were addressed. At the time outliers were typically games requiring drivers and firmware patch.


    DEC Alpha had x86 software binary translation, in 1993, called FXi32. Overhead was enough to make FX!32 a non contender until Alpha NT 4.0 native.


    Nx586 was the first unique and differentiated x86 RISC design to successfully implement x86 CISC instructions by RISC decoder in silicon.


    Maybe ARM 64 hardware performance, on leading process, will be robust enough not to notice translation overhead; which Elbrus proposes on their simulations.


    Or one of the licensees has implemented a hardware translator, x86 to RISC native decode, in silicon.


    To process some x86 instructions traditionally means Windows, or any processing engine supporting x86 RISC Ops, targets simply a sub set of the total x86 instructions. Those specifically required for the target application.


    That's the design secret. Binary translation does not have too address every x86 instruction. Just those needed for the target application.


    This also depends on the performance of simulators too perform top down model compatibility verification.


    FB saying in El Reg, "It's also been crucial in our efforts to get hhvm running on ARM processors by isolating and reducing the amount of architecture-specific code we need to re-implement,"


    Where the key FB statement is "crucial in our efforts to get hhvm running on ARM processors by ISOLATING AND REDUCING the amount old architecture specific code we need to re-implement”


    If such a suspicion is any where near accurate, Intel network will have secured the simulation, or carted off with a decoder, and will certainly have engineers and attorneys exaiming this unique and differentiated ARM implementation closely.


    Mike Bruzzone,
    Camp Marketing
    15 Dec 2013, 10:04 PM Reply Like
  • Mike Bruzzone
    , contributor
    Comments (127) | Send Message
    Just to make sure everyone gets the legal position correct in relation to Intel IP Engineering Attorney network, "a wholly unique and original implementation of the x86 compatible instruction set architecture, or, x86 compatible instruction set microprocessor".


    Intel attorneys hate the description "wholly unique and original implementation of the Intel x86 standard compatible microprocessor".


    Mike Bruzzone
    Camp Marketing
    15 Dec 2013, 10:54 PM Reply Like
DJIA (DIA) S&P 500 (SPY)
ETF Tools
Find the right ETFs for your portfolio:
Seeking Alpha's new ETF Hub
ETF Investment Guide:
Table of Contents | One Page Summary
Read about different ETF Asset Classes:
ETF Selector

Next headline on your portfolio: