Original URL: http://www.theregister.co.uk/2013/05/20/intel_chip_customization/

Intel's answer to ARM: Customisable x86 chips with HIDDEN POWERS

Let's all play find the secret hardware register

By Timothy Prickett Morgan

Posted in Servers, 20th May 2013 10:06 GMT

With new CEO Brian Krzanich and new president Renée James in control of Intel, all kinds of changes are very likely in store: the chip giant wants to expand beyond its dominance in PCs (a declining market) and servers (one that is profitable but not growing very much) to other aspects of the computing landscape.

And one such area that Intel has been dabbling in - and will very likely jump into to compete against the ARM collective - is the customisation of its x86 processors for specific large-scale customers in the server racket.

This is not at all a surprise to anyone.

Everyone expects any processor package maker to do deep sorts through their chip bins, finding components that can handle slightly higher temperatures or that have their voltages and clock speeds altered to fit in a specific thermal envelope, or that can deliver higher performance as clocks are cranked above the thermal limits of standard processors.

(Over in the ARM world, this has been going on for years and far more comprehensively: the British processor core designer flogs its technology blueprints to Samsung, Qualcomm, Broadcom et al, who package the CPUs with all manner of electronics to build heavily customised chips for specific applications.)

As it turns out, Intel has been doing more than just picking out devices with unusual temperature, power supply and over-clocking tolerances, and is in some cases actually etching different features or instructions onto its silicon for specific customers. This is a step outside the chip giant's comfort zone.

"We are definitely doing all of the above," confirmed Jason Waxman, general manager of Intel's Cloud Infrastructure Group.

"As far as the etching goes, we have done different things for different customers, and we have put different things into the silicon, such as adding instructions or pins or signals for logic for them. The difference is that it goes into all of the silicon for that product. And so the way that you do it is somebody gives you a feature, and they say, 'Hey, can you get this into the product?' You can't do something that takes up a huge amount of die, but you can do an instruction, you can do a signal, you can do a number of things that are logic-related."

In many cases, after the customizations were done and features were added to x86 chips, these features are not explained in the documentation and are only available to the customers who know the feature exists. (Such practices are not rare in the world of embedded electronics, weary hardware engineers will remind you.)

"In some cases, we have created new SKUs or changed the parameters based on the data center operating requirements if we know the exact temperature and the exact system design," continued Waxman. "Based on their application, their system, and their data center, I can come up with a unique combination of cores, frequency, power, and et cetera that really targets their sweet spot and their particular needs."

Intel is not, at this stage, doing custom blocks of logic for x86 processors where the chips for customer A and customer B are different from the stock chips that everyone gets. But this could be a way for Intel to leverage its chip design and manufacturing prowess to take on the ARM collective, which has dozens of companies hacking the core ARM Holdings designs for very specific uses.

'Everybody hates the golden screwdriver upgrade approach'

"We are going down that path, and more and more with SoCs [system on chips]. But today the silicon customization comes in the form of designing a SKU based on their unique requirements or adding logic that will go into the mainstream product line but activated by a unique customer."

Waxman cannot be specific about precisely what kind of logic tweaks it has put in, but they are generally instructions to accelerate a function, or maybe extra management features, or different signals on a pin.

Hidden features and latent capabilities in x86 chips that only select customers are aware of is sure to annoy some server makers and server buyers, who will now want to know what these chip tweaks are and how they can access them.

Everybody hates the golden screwdriver upgrade approach, where a feature is either hidden or activated through software, but the truth of the matter is that chip makers have been doing this sort of thing for decades – and charging extra for it.

Intel is not talking about the precise business – meaning money – that comes into play when these customizations are done on its various processors. But it stands to reason that sometimes the changes are necessary just to get the chip sales and are therefore worth doing in their own right. But in the future, such customization could end up generating incremental revenues for Intel, or at the very least, help it to recoup its engineering costs as chip and system-on-chip customizations do not map as easily to mainstream products.

"We are moving rapidly in the direction of realizing that people want unique things and they are going to want them in silicon. In some cases, it will be done in software," said Waxman.

"There are SoCs and multichip modules. Sometimes a monolithic die won't be the right approach and customers will want multiple chips on a package. Having a CEO that is a manufacturing guy certainly comes in handy. We are definitely getting more requests for customization and we know we have to be best in class at it."

At the moment, though, this customization has been restricted to about a dozen big server manufacturers and hyperscale data center operators, who have the same scope as many server makers if you look at the number of boxes they consume.

Intel has a one to two node chip manufacturing process lead over its rivals in both the x86 and ARM processor markets, and both GlobalFoundries and Taiwan Semiconductor Manufacturing Corp know they have to try to close that gap to compete against Chipzilla.

Intel has a broad product line already, with its Atom, Core, and Xeon chips – yes, Itanium is in there, but it is essentially a custom processor made for Hewlett-Packard at this point – and it will have to be careful to allow enough customization to allow its x86 chips to retain their hegemony in the markets where they dominate – PCs and servers – and to expand into new markets – microservers, smartphones, and tablets – while at the same time not overburdening Intel with chip engineering and manufacturing costs.

The combination of flexible chip design, manufacturing process lead, and deep expertise with infrastructure and application software, which Intel is building up, could level the playing field with the ARM collective. Provided, of course, that the costs are not too high.

The one thing that Intel almost certainly is not going to do is make ARM processors in its fabs for third parties. ®