The Intel-ARM power struggle

Almost a decade ago, back in 2008, ARM was battling Intel in the mobile processor space using its three advantages: as the incumbent mobile processor leader, as a licensor of cores rather than a supplier of  single-sourced chips, and as the power consumption leader.

As the power consumption leader, ARM always though Intel would catch up one day, but never knew quite when.

“How incredibly fortunate it is”, Warren East, then the CEO of ARM, now chairman of Rolls-Royce, told EW in July 2008, “that Intel has done a job that we would consider not a very good job on that.”

According to ARM’s then vice president for marketing, Ian Drew, who was formerly an Intel executive: “We are one or two orders of magnitude better on passive power than Intel.”

Drew was supported, at that time,  by the wireless industry’s leading analyst, Will Strauss of Forward Concepts, who points out that the ARM-based products of Qualcomm and Texas Instruments are ten times or more lower in power consumption than Intel’s mobile processor Atom.

With Intel having had an ARM architecture licence since 1997, and produced ARM-based processors, which it called X-Scale why hasn’t it learned to do low-power?

“It’s not one big thing,” replied Drew, “it’s lots of little things. It takes years of experience. But I’m sure they’re trying to do it.”

A massive advantage for ARM is all the licensees who have spent years designing low-power subsystems and peripherals around ARM cores.

“It’s not just the microprocessor but the system around the microprocessor”, said East, “we have upwards of ten years experience of building power-efficient systems on chip with multi-innovative sources in the dozen or so large semiconductor companies which have been developing innovation to ARM-based systems.”

Intel says it will reduce the power of Atom, by an order of magnitude in its next generation, expected in 2010.

But, by then, ARM would also have moved on. “There tends to be a 50 per cent improvement in power efficiency at each new generation of ARM,” said East.

TI, which used an ARM-based core in its OMAP platform to compete with Intel’s Atom, reckoned OMAP uses five times less power than Intel’s Atom.

According to TI’s wireless boss in 2008, Greg Delagi, even with an order of magnitude reduction in Atom’s power dissipation, Intel would be “still not in the ballpark of where they need to go.”

 


Comments

18 comments

  1. I would have assumed they wouldn’t take the lead on it Craig, but they’d adopt it if someone else did.

  2. Craig Abernethie

    Would it be possible to get the big capacity firms to make a consortium to make Supercomputers for the Arm Architecture? Approach Fujitsu or Cray to make the SC for as they are famous for doing that type of work. I don’t know how many cores a basic SC machine would need but i’m sure it would be at least 10’s of Millions or even up to a Billion cores for a record setting machine. So up to 250 Million Quad-Core chips for some type of Giga-Computer!!! It would be great to see. Thank you all for listening.

  3. Yes indeed SEPAM, Intel is excellent at picking the wrong future markets to get into – watches, Wimax, mobile, wearables, jewellery and, I suspect, IoT and driverless cars.

  4. I don’t fault the aspiration, Paul K, just the execution. As with X-Scale, Intel never put its Atom-based mobile SoCs on in-house leading-edge processes.

  5. Spot on DontAgree. It’s interesting that Intel knew all of that and still thought it was worth having a go. Must have have cost them $10bn+.

    • Yes, they were spending a $B/quarter paying companies to use Atom in phones and tablets, so they spent a lot. But, you cannot fault them for trying as they saw the risk of many people going to phone/tablet only vs. PC, which is their core market. Chromebook is another risk example, where you have many people who really only use PCs for browsing, email/messaging, and writing some docs – Chromebooks work in x86 and ARM, so blurs lines.

  6. Yes indeed, Paul K, trying to usurp an incumbent micropocessor architecture is next door to impossible – Intel found that out to their enormous cost in mobile and ARM is finding it out in the data-centre,

    • And the strength of that incumbency has a lot to do with ECO system and marketing experience.

      Indeed you are only going to up-end the market if you have a significantly differentiating benefit over the incumbent that is worth all the extra cost associated with ECO system overhaul (and know how to sell said benefit in that market).

    • SecretEuroPatentAgentMan

      Intel dominates the desktop and practically controls the server market.

      They have already seen ARM now practically owns the embedded market. So why then did they not look for new emerging markets rather than trying to get into embedded by way of extreme bribes? Rather it seems Google has realised this and is now filling their server rooms with their own TPU, now in the second generation already, and also Lanai which we know less about. For all we know they might next sell these processors and they have even provided Lanai related patches for GCC which would be odd if they are not providing the chip to outside parties.

      The DSP market is fragmented but Imaging Signal Processors seem to be a new field that is up and coming with notable absence from Intel.

  7. A server chip and a mobile chip are completely unrelated beasts. Mobile chips have quite varying loads and have to be able to go into very low power states and come out of them quickly to save battery but also service wireless connections. ARM was trying to pick off the lower end of the servers such as SOHO and cloud servers, but Intel was able to answer that quickly with an Atom based server, while still maintaining all of their experience in servers.
    In essence, incumbency has a lot of value..

  8. No Pedro, I was talking about market share and you were talking about performance per watt. I think your confusion derives from the 2008 performance which the post is all about and the 2016 performance when Intel gave up in mobile CPUs. Atom did eventually catch up ARM in performance but even Intel couldn’t stay in a market where it was losing $4bn a year.

  9. Once Intel caught up with ARM in power efficiency, the only advantages ARM had were its independent licensing model and its incumbency IMHO, SEPAM. Now that it has surrendered its independence it has only its position as the incumbent market leader in mobile CPU to sustain it. But that won’t help ARM
    in other areas like servers where, as you say, people like Broadcom and Marvell have given up making ARM-based server chips, where Qualcomm is flurting with RISC-V and where AMD has gone quiet on Seattle – its ARM server chip project.

  10. Yes I do indeed think it, Pedro, ARM CPUs are in a lot more mobile slots than Intel CPUs. QED.

    • I see what you did there… you confused “performance per watt” with “market share” .P

      • SecretEuroPatentAgentMan

        Those are related by the consumers negative view of battery guzzling phones, thus directly impacting the market shares.

        Intel was late to efficiency but is more well known for raw performance. Nevertheless it would be interesting to see how RISC-V and AndesCore perform and I see some SOC providers including Qualcomm have also shown interest.

  11. Well you see Pedro every Friday afternoon the site carries a ‘Yarn’ – an industry story from the past. It can be about anything. The Intel-ARM struggle for mobile was a not uninteresting saga IMHO. As for the conclusion, in 2008, when the story was written, we didn’t know it but now, in 2017, we do know the conclusion – ARM won.

  12. David, why does this article exist and what is its conclusion ??

    It consists of a series of unconnected quotes from a decade ago with the constant theme of “X86 is power hungry beast”. When in reality all the evidence shows that there is no X86 power penalty despite what ARM’s marketing department would like he world to believe.

    Otherwise how do all the ARM server projects lose badly in efficiency and how come it is (was) possible to make X86 based phones with reasonable battery life (zenfone 2, razr-i).

Leave a Reply

Your email address will not be published. Required fields are marked *

*