PORTLAND, Ore. – What with AMD, Apple, Dell, Dialog, Freescale, Fujitsu, HP, LSI, Microsoft, Motorola, MStar, Nvidia, Qualcomm, Samsung, STMicroelectronics and Texas Instruments all licensing ARM cores for everything from smartphones to tablets to basestations to servers, one might be led to believe the boast of ARM CEO Warren East that designers are choosing ARM because it is a no brainer.
The truth is that even though ARM is gaining market share in low-power embedded applications the company still has a long way to go to challenge the perennial microprocessor kings; that software compatibility hurts ARM as much as it helps; and that there are many vertical markets which will remain inaccessible to ARM cores merely because they are standardized.
Intel in particular has a 20-year head start over ARM, resulting in a maturity, sophistication and veneration that will be hard to displace by a 22-year-old. Intel's ecosystem of support chips, subsystems and software is unparalleled in the industry and addresses many more real-world issues than the low-power and small die size that makes ARM a no-brainer for many new designs. From the mobile space where Atom offers x86 compatibility that even ARM's most sophisticated cores cannot match, to the server space where Intel's Xeon already solves the most vexing issues facing datacenters today, Intel versus ARM remains a brain-teaser. For instance, the fast-growing cloud computing space uses virtualization to offer mobile device users access to applications running on servers for which Intel has a top-to-bottom solution – VTx – which securely links x86-based mobile devices with Xeon-based cloud computers. ARM, on the other hand, is still pursuing virtualization extensions that could offer similar integration of mobile devices into cloud computing realms.
"ARM is becoming more and more mature – its graduated from its initial light-weight processors for mobile devices to beefier processors like the A15 which can compete with Intel's Atom, but they are not yet in the Xeon class" said Sergis Mushell, principal research analyst at Gartner Inc. "The battle has just started – the relay gun has been fired, but the race has just begun."
ARM has recently started penetrating the wireless basestation market with a vengeance, driving a stake in the heart of MIPS Technologies, and landing lucrative new accounts such as for Freescale's popular QorIQ basestation processors.
However, standardizing on ARM cores may eventually hurt margins at TI, Freesale, LSI Logic and elsewhere, as they compete for the same sockets since one of ARM's advantages – interchangeable software – will also enable processor vendors to become interchangeable.
"ARM is now threatening the ecosystem of MIPS – going into basestations, going into networking, and gaining mainstream licensees like Texas Instruments, Freescale and LSI," said Mushell. "But all these vendors are going to be competing for the same piece of business, and it will be really easy to replace one vendor's ARM processor for one of the others, which could drive the ASP [average selling price] down significantly."
For many applications, proprietary architectures can actually be an advantage – witness HP's recent decision not to support ARM on Windows 8 tablets. Proprietary cores can also serve specific markets needs that do not fit into ARM's one-size-fit-all philosophy. For instance, Infineon's Tricore, Renesas' VH/RX and Atmel's AVR all serve specific market segments that will not be going away any time soon.
"Though it's true that the ARM ecosystem for embedded microprocessors and microcontrollers is growing at a blistering rate, there are applications – such as in automotive microcontrollers or networking microprocessors, where manufacturers have found it useful to maintain proprietary architectures," said Adib Ghubril, research director at Gartner.
The other dimension that ARM's "no brainer" argument is skirting is heterogeneous multi-cores. ARM currently has extensions aimed at addressing digital signal processing (DSP), vector processing and other specialized needs, but at least for now, chip designers are more likely be use an asymmetric architectures – an ARM core in combination with a non-ARM heterogeneous core – to meets their specific needs.
"And a new war is about to begin – the multicore wars – and here ARM will likely not emerge supremely victorious, because devices are taking on what we refer to as asymmetric architectures to meet requirements specific to end-applications," said Ghubril. "These asymmetric architectures combine heterogeneous cores together such that an ARM may be sitting next to a proprietary DSP."
I have to agree this is a Intel sponsored article. Almost every MCU company is licensing ARM core nowadays. Toolchains can be easily obtained from Internet and it is relatively easy to kickstart development for ARM.
More ARM cores are produced in a single quarter than all the Pentiums Intel produced ever.
Revenue is a poor measure of market share.
Rather look at design wins. Where are x86 used? PCs and a few other application areas. ARM, OTOH, are everywhere - including in that PC.
There are more ARM cores in a laptop than Intel cores. The same likely goes for servers too.
I don't agree that heterogeneous CPU choices is a good idea - well for embedded anyway.
Twelve or so years ago I worked on 5 different projects in one year - each had a different CPU architecture with its own quirks, tools, debugging equipment. Not at all productive.
These days I tend to work with ARM most of the time. One toolchain, debugger and architecture covers everything from sub-dollar micros to GHz multi-cores. No need to keep learning new architectures for every job!
"Small player in the real world".
How are you measuring that? Which numbers are you using?
Pick the wrong numbers and you get the wrong story.
For example, a Hummer is more fuel efficient than a Prius. Did you know that? Pound-for-pound a Hummer produces better mpg than a Prius!
If you rate market share in dollars you fall in exactly the same trap.
In my real world there are far more ARM-based devices than Intels. Only my PCs (8 or so of them) have Intels.
Even in your PC there are likely more ARMs than Intel CPUs. Your disk drive has 2 or 3 ARMs, the Bluetooth and Wifi modules each have one...
Intel has no interest in playing where ARM currently plays. Intel only understands selling high margin parts. They do not understand how to do what ARM does and sell IP to other people actually making the parts.
"I don't really see how there remains a market for Atmel AVRs, apart from supporting legacy products." - Then you did not look very deeply.
Does that ST part come in wide supply ? oops, no.
Does it come in low pin count packages ? oops, no.
That's two large market determinants right there.
Wow. Bad info all around. Starts with the 12 years comment. Try early 1980's. VLSI Technology. But the article dateline is Portland - not far from an Intel stronghold. Hmmm.
No doubt: Intel makes big bucks designing, building, and selling screaming microprocessors. But don't underestimate the significance of ARM cores. Intel doesn't.
We need both of these architectures (and a few more). ARM is a very different kind of company than Intel. Again, we need both (and many more).
Lots of issues in this article that are twisted badly.
as short history lesson for R Colin and an insight in to the latest ARM Cortex-M0 core is presented by no other than Mike Muller one of the founders of ARM
in his KEYNOTE: General Session/Awards and Scaling for 2020 Solutions video here
you have to sit there and wait for it to fully buffer so you can skip
the the start of his presentation at 28 minutes 30 seconds and there's no easy way to download the clip here but watch it anyway, its insightful and informative.
its a shame they used this private service to show the video publicly rather than the simple googl+ "hangouts" video and so have it all automatically recorded to YouTube in HD for easy reference later.
as make says do the numbers :)