Although the x86 architecture is as dominant as ever in the server and desktop space, there’s a small number of people rooting for ARM chips to move in on the territory.
Venture capital funding has found its way to them, so they might not be as crazy as they seem. Dell admitted last year that it also has been playing with ARM chips in server configurations. Even Google might be getting in on the act; last year it purchased Agnilux, a tech startup that was rumoured to have been designing an ARM-based server.
Right now ARM chips aren’t found much outside of phones and tablets, although they’re doing very nicely in that regard. They power the vast majority of mobile devices around the world, and have done so for some years. ARM doesn’t make chips itself but instead licenses the design to manufacturers like Samsung, Texas Instruments and Qualcomm.
However, the goal is to produce computers that draw significantly less power compared to x86 offerings, making them massively cheaper to run. Manufacturers talk of power-per-watt figures, and here ARM chips win without even breaking a sweat.
Intel is feeling the pressure. Because many companies manufacture ARM chips, it’s under attack from several different directions at once. Earlier this year Nvidia announced it was creating an ARM desktop/server chip, for example, and late last year Marvell announced its plans to produce a server chip.
The latest company to throw its hat in the ring is Calxeda (I don’t know how to pronounce that either; they were formerly known as Smoothstone). Its chips will be based on the ARM Cortex A9 quad-core design but–like the SeaMicro SM10000-64 low-power server announced last week, which uses Intel Atom chips–the goal is to make a chip that can be packed into a server in multiples of up to 120.
And in the short term this might be the trend. Yes, ARM might have an impact upon the server space but only in highly-specialized high-powered areas, where ponying up a few hundred thousand dollars for hardware isn’t unusual.
Yet in some ways the signs are right for a potential explosion in server–and perhaps even desktop–ARM systems. There’s a variety of manufacturers out there, making for a competitive marketplace that could keep prices low. And at the 1-2GHz speeds and dual/quad-core architectures, ARM chips have arguably reached a baseline level of performance that’s good enough for a variety of tasks, especially if several chips are combined on one motherboard. (Remember that heat dissipation just isn’t a huge issue with these low-powered chips, so there are no design issues.)
But you shouldn’t hold your breath for an ARM server or desktop near you soon.
We have the chips, but what about the rest of the hardware? ARM chips use the same DDR memory and disks as regular computers, but check your nearest PC supplier and you’ll find no motherboards that support ARM.
Unlike the well-established Intel and AMD marketplace, where it might be possible for manufacturers to create boards based on reference designs, ARM server or desktop manufacturers are going to have to engineer their own designs. That’s not going to be cheap. There are a handful of ARM development boards on sale, as well as some designs for embedded use, but these are intended to ease creation of software for embedded systems.
On the other hand, the system-on-a-chip (SoC) approach that ARM chips use means that some of the necessary chipset logic is contained in the chip itself, reducing costs elsewhere. However, if two or more ARM chips are going to be utilised on one motherboard–which is perhaps the most sensible plan–then some kind of logic controller will be needed to let everything intercommunicate. This will have to be expensive custom silicon.
Then there’s the software. Microsoft is in the process of porting its Windows architecture over to ARM chips, but that seems to be more about bringing Windows to mobile devices than elevating ARM to the same status as the x86 platform. Dell has also gone on record saying the software ecosystem for ARM is weak and that porting software could be expensive, leading to organizations maintaining two separate software stacks: one for x86, and one for ARM. That won’t be cheap.
None of this is helped by the fact that ARM has yet to produce an entirely 64-bit design; all ARM chips are presently 32-bit internally. There’s seemingly no intention to change this situation either, with ARM’s CEO recently suggesting that 32-bit ought to be good enough for everybody. Most server admins consider 64-bit as de rigueur and simply won’t accept anything else, if only because it’s hard (and possibly even embarrassing) to make a funding proposal for a 32-bit setup in this day and age.
But, but, but… These are still very early days. We can’t blame ARM for focusing on its core competencies right now. And although ARM might be headed for niche high-end servers right now, it’s possible that a trickle-down effect will occur whereby the confidence engendered will embolden others to start producing ARM-based blade servers, or even desktop computer.
Keir Thomas has been making known his opinion about computing matters since the last century, and more recently has written several best-selling books. You can learn more about him at http://keirthomas.com. His Twitter feed is @keirthomas.