3DVelocity logo


Content

Latest

News
Articles

Community

Forum


Main
If ARM's official launch of its Cortex A50-series of processors was set to make Intel quiver in its snake skin lined boots then the fact that chips based on the ARMv8 architecture won't see the light of day until 2014 will hand the advantage back to Intel.

ARM's A50-series of chip designs are not only intended for servers, the firm expects it to end up in tablets too and touted its Big Little architecture. For ARM its success will be gauged on how Cortex A50 variants do in the server market because the firm has very little market share to gain in the tablet market.

ARM's problem in the server market are two fold, software and Intel. The firm is hoping to replicate the astounding success Intel managed with its Xeon processors where as of 2012 Intel commanded a 94.3 percent marketshare according to IDC, but unlike Intel it doesn't have a slow, overpriced incumbent to deal with. Instead it is facing a company that is aggressively chasing down ARM's only competitive advantage and could well bridge the power gap before ARM's licensees get a chance to make a serious dent in the server market.

Xeon ignites an inert market

Intel's Xeon processor was launched in 1998 replacing the firm's first real foray into the workstation and server market, the Pentium Pro. Within a decade it wasn't just owning the market, it was hard to remember a time when servers were running a UNIX variant on RISC processors. The feat becomes even more astounding when you realise Intel didn't release a dual-socket part, codenamed Foster, until 2001.

Intel managed this through one simple tactic, it undercut the UNIX server vendors considerably. The firm also offered the chance for system builders to get into the server market without having to invest heavily in research and development, meaning there were far more firms pushing Intel-based server kit than ones running Power, SPARC, MIPS, Alpha or some other, what would now be called exotic, ISA.

Intel was helped by two things, UNIX vendors that looked down on Intel's architecture classing it as consumer-level hardware alongside the operating systems that ran on the IA-32 architecture and the first dot-com boom. Traditional UNIX vendors such as IBM, HP, SGI and Unisys wanted to continue pushing high margin systems that could not be easily replicated by rivals. HP and IBM may have been selling desktops and laptops with 'Intel Inside' but at that time any Xeons found inside HP and IBM's product ranges were merely paying lip service to a vendor that weilded a disproportionate amount of power.

The most graphic illustration of distain for Intel's IA-32 architecture came when HP and Intel got together to work on IA-64, the ISA behind the oft-lamented Itanium processor. Thanks to Oracle, we now know Intel wanted to get rid of the Itanium but due to contractual obligations with HP, it has to continue developing and supplying HP with Itanium chips until HP decides it has had enough.

Whenever I talk to HP's mission critical folk, I still hear someone from the firm claiming its Itanium-based Integrity and Superdome servers have hardware features that more common Intel servers including its own units, lack, but that argument has become largely academic thanks to virtualisation and 'the cloud'. The truth is, the big monolithic server that has to be always on no matter what is finished as a viable business save for one or two vendors that will make do feeding off the scraps left by firms that won't change due to ultra conservative business practices something Kuda Stolariski, research manager of enterprise servers at IDC believes stating on 28 August 2012 that, "IDC expects the Unix market to stabilize over the next few years and remain a smaller, specialized segment of the overall server market, catering primarily to customers who are looking for high levels of availability and stability in a scale-up architecture."

Intel was also in time to cash in on the dot-com boom, a time when new companies that were not burdened with decades of buying behaviour favouring Big Blue, desperately needed cut price servers to power their growth. Quite frankly, Intel's timing and market positioning was perfect and the analyst market share figures confirm this but perhaps the fact that just about every UNIX vendor now relies on Intel-shod servers to bring home the bacon, is an even bigger accolade.

While Intel was able to play the price card to win business, ARM is betting on energy efficiency, which everyone argues is ultimately price in the form of total cost of ownership (TCO). But relying on TCO is a dangerous game because not only is it opaque, after all energy prices vary, many chip designers I've spoken to including ones at ARM's latest buddy AMD, repeatedly say there's nothing 'low power' about ARM. There will always be a sense of bias in such a statement and quite frankly it doesn't matter at the ISA level, what really matters is at the plug and there is no doubt that Intel has dragging its X86 architecture down to the power levels it needs to compete with ARM in smartphones, tablets and servers.

Processing away ARM's power advantage

Intel's biggest strength now is not its X86 architecture or the incremental developments it makes but its manufacturing capabilities. It is at least a generation ahead of its competitors, namely GlobalFoundries, TSMC and Samsung and although it does offer capacity to other chip vendors, more than one person from Intel has told me that it won't sell capacity to anyone that it believes is creating a rival product. Intel's look but don't touch strategy makes business sense all around, open your doors to appease regulators in case they start sniffing around but only offer the good stuff to those it doesn't deem a threat.

Intel's Medfield Atom chip has done surprisingly well despite being showcased in middling smartphones. It still surprises me why Intel didn't use the Motorola Razr I as the first Medfield-based smartphone rather than relying on the reference design used for the Orange San Diego. Most journalists I've spoken to that have used the Motorola Razr I have said they were surprised with the battery life of the phone, something 12 months ago everyone thought would be Intel's Achilles' heel.

Aside from Medfield, Intel is also bringing its full-fat Core architecture down to the 10W TDP market in the coming year. Dadi Perlmutter, executive vice president and general manager of Intel's Architecture Group, said at IDF the firm expects 10W TDP Haswell parts to be some of the most popular. Haswell will be based on the rapidly maturing 22nm process node that made its debut with Ivy Bridge and come 2014, Intel would have already got its 14nm process node ready for production according to Mark Bohr, director of process architecture and integration at Intel.

All of this means that while ARM's vendors are getting ready to lift the lid on their ARMv8 processors in 2014, Intel will have X86 chips running below 10W and though that doesn't taking into consideration system-wide power utilisation and overall performance in actual real world workloads, both of which are vital for commercial success. It is Intel's performance on known workloads that may give it a nod in datacenters even if its chips have TDPs that are a few watts higher than ARM-based rivals, something AMD has faced and lost out to in the past decade.

One interesting part of ARM's attempt to muscle in on Intel's turf is that it will either validate Intel's decision to invest so heavily in fabrication plants or show that fabless chip firms such as Nvidia and Qualcomm can stick it to Intel without the need to invest upwards of $5 billion on a single fabrication plant that will need retooling every three to four years if Intel continues down its tick-tock roadmap.

So while Intel looks on track for 14nm chips to appear in 2014, the proof is if the firm can continue to cut power in successive generations in a timely and economically viable manner. I've asked a number of Intel staffers, including Bohr and Justin Rattner, CTO of Intel, whether the firm can continue to afford to continue following Moore's Law and the answer from both is essentially the same, for now Intel can but not forever. Even Gordon Moore has publicly stated his 'law' won't continue forever and economics will judge when Intel has to finally give up on what is now largely a marketing term.

In the short term Intel has both a process node that will bear fruit just in time to counter ARM's licensees and as I write in my next article, a significant software advantage too. ARM's battle, despite all the hype, hasn't even begun.

comments powered by Disqus