Intel vs. ARM

“Intel vs. Arm” returns 65,000 hits from Google, and in general much is made of the contrast between Intel’s near-dominance in the high-performance market and ARM’s near-dominance in the phone/tablet market. But it seems that:

  • Intel’s important asset is its massive fab capacity. In a web-based and mobile world, their aged ISA is increasingly irrelevant.
  • ARM has no fabs. They do, on the other hand, have great processor designs for the medium-performance, high efficiency space.

This doesn’t sound to me like a battle of the titans. It sounds like a wonderful partnership waiting to happen. What am I missing? Of course I am aware of Intel’s ARM license, but that just makes this more mysterious.

13 Replies to “Intel vs. ARM”

  1. It strikes me you’re underestimating Intel. It’s true that no one much cares for x86, but the performance penalty you pay for using it is small and fixed (i.e. getting smaller as processor scales in size and speed.) Intel _could_ issue the next set of processors using the ARM ISA, maybe that’d even be better for everyone…but it doesn’t much matter for almost anyone.

    What is true is that in the high performance world, it isn’t just Intel’s fab that matters; they also seem to have the best designs around (in many categories; graphics probably not being one of them.) Even at a fixed # of transistors, process, and power budget (if that power budget is more than 50W :P), they do a much better job than most others at getting the last few percent out of their hardware. AMD is learning to their dismay that competing with Intel, head-on, is really, really hard.

    That said I think that you’re right that Intel & ARM should work together. Just don’t count out Intel’s designers. They know what they’re doing.

  2. Hi Andrew, I agree about Intel’s engineering prowess.

    I’m just wondering why Intel’s management would run (what seems to me) a very real risk of losing a war, when this war doesn’t actually need to be fought.

  3. Just to elaborate a bit: this seems like a classic innovator’s dilemma for Intel.

    The computer industry has no shortage of examples where companies filled with great engineers became irrelevant because they engineered the wrong products.

  4. Intel have for years tried, without much success, to enter the ARM-dominated low-power/mobile market with its Atom line. Although the more recent Atom chips are less of a joke than the first ones were, they are still no match for ARM designs. ARM, meanwhile, has been slowly moving towards increasingly higher performance without sacrificing the low power consumption. The currently most powerful ARM core, the Cortex-A15, has performance and features (virtualisation, large physical address space) making it a viable choice for a server platform, and companies like Calxeda are offering such systems.

    With energy increasingly becoming a major cost for data centres, ARM-based servers may have a chance of gaining some real market share, and this is should be worrying Intel. For the first time, ARM and Intel target markets are without doubt overlapping, even though neither poses a threat the other’s core business.

    Intel does of course have the option of admitting defeat and taking up manufacture of ARM chips. Profits from these chips would, however, be (mostly) in manufacturing margins, clearly a less attractive alternative than selling their own designs. Furthermore, such chips could help ARM gain a foothold in the server market, meaning Intel would have to fight to retake, not only to keep, this segment to themselves (well, and AMD). The fact that Intel has an ARM architecture licence is largely irrelevant to any of this; it merely adds a little irony to the situation.

    On the consumer side, cheap Atom-based netbooks running Microsoft Windows have been popular for a while. However, the increasing number of tablets on the market as well as the imminent arrival of Windows 8 with ARM support might shake that up a bit.

    Finally, it is interesting to compare the sizes of the two companies, ARM employing about 2000 people and Intel almost 100000.

  5. If it could, Intel would certainly buy ARM, kill the two dozen or so ARM processor makers, extend its monopoly to low power and quadruple the prices and have everyone buy from Intel. The ARM ecosystem is wonderfully efficient from the fact that is not dominated by a vertical. There is competition on many levels. ARM minimally contributes only an instruction set and demands a very reasonable tax for it. The rest of the creativity is going on in the ecosystem which has no wish to see their very good friend ARM – which is simply a standard if you like – to be replaced by the Intel bully. Intel buying ARM would kill ARM instantly. I think MIPS would be first in line to replace it.

  6. Andrew: Yes, using more smaller cores is only feasible if each one is “fast enough”. There is every reason to believe the Cortex-A15 is fast enough for many applications which are anyway highly distributed and/or mostly I/O-bound. It is of course not a contender for applications where CPU-bound single-thread latency is critical.

    Regarding peripheral costs, there are savings to be made there as well. The lower power allows these CPUs to be packed more densely (Calxeda claims to pack 480 cores in 2U), simplifying network and power supply infrastructure.

  7. I find Mans’ comment interesting. Allowing for the fact that Intel own fabs and will obviously need to staff them, does anyone know how many of Intel staff do equivalent work to ARM, i.e. design and managing partnerships?

    From what I know, ARM never set out to compete with Intel, they identified the need for a strong CPU design which could be licensed and used as part of a SoC. Simply, ARM grew the product range bottom-up whilst Intel have grown their range more or less top-down. The Atom is a scaled down Pentium.

  8. Irrelevant ISA?
    Not so much: why do you think Intel continues to makes only x86 when they could also produce ARM CPUs?
    Because with the x86 ISA, they have only limited competition whereas with the ARM ISA all the other fabs are competitor..
    Paranoid Intel prefers to avoid competition but using the ISA lock-in even though they are probably the bests for CPU making.

  9. Eugene: Absolutely but that’s why Intel are facing the Innovators’ Dilemma: their bottom end is being disrupted as ARM cores become more and more viable for the majority of everyday computing.
    The Dilemma is whether they commoditize their own bottom end and sacrifice margin or whether they wait for someone else to do it and get pushed further and further up the market, eventually becoming irrelevant.
    The latter is exactly what happened to all the Workstation vendors in the mid-90s.

  10. Reno, I think the ease with which most computation has moved to x86-64 shows that x86 lockin is no longer a significant factor. My Macbook, my Linux machine at work, and my wife’s Windows 7 machine all have some x86 binaries, but not very many.

  11. Intel had an ARM architecture license, and was doing Ok with XScale for a while. What really hurt them is, when push came to shove, Intel’s moneymaker was the PC side. So any time there was any conflict, the PC side won at the expense of the mobile side. While that’s probably the right strategy for Intel short-term, it ended up stunting the growth of their XScale team. Why would Intel make mobile chips with low margins when instead they can make desktop chips with high margins? This is also why I expect Intel’s foundry service to quietly disappear.
    Also, they had such hubris within their Atom team on power consumption that they felt they didn’t need the ARM architecture, and also that it would be easy for developers to support a PC ecosystem in mobile, ie run the same apps on the mobile and desktop platforms. Reality is that the two platforms are not very alike, and so you’re unlikely to ever run the same app on both (with the same expectations anyways).
    Perhaps Intel’s purchase of Infineon means they understand what they did wrong and are trying to catch up? Most likely not. My prediction is that they will instead try to glue a baseband to an Atom and wonder why tablet and handset manufacturers don’t want to use their expensive, power hungry device.

  12. regehr: The x86_64 transition went smoothly because every x86_64 CPU can run legacy 32-bit code (and 16-bit) allowing PC manufacturers to start using the new chips even though software was not yet widely available. A 64-bit OS can run 32-bit applications enabling a gradual transition rather than forcing users to replace all their software at once. In fact, most Windows installations are still 32-bit, almost a decade after the first x86_64 CPU saw light of day.

    Apple was able to change the CPU architecture of the Mac twice (m68k to ppc, then to x86) only by providing comprehensive emulation of the older CPU with the new systems, and this was feasible only with the new CPU being significantly faster than the old one. They also allow creation of “fat” binaries containing code for multiple architectures hiding such details from the users who would otherwise have to choose the correct package for their machine. Neither of these approaches are particularly attractive for a resource-constrained, low-end device, which is where any x86 to ARM transition will happen.

Comments are closed.