Yesterday I returned from the Intel Developer Forum to be greeted by some interesting news that definitely wasn’t mentioned at IDF. In case you missed it, the CPU Goliath has started selling upgrades to its low-end Pentium CPUs in the form of scratch-cards – in the US, at least.

It’s a simple idea. The processor ships at a low price with some of its capabilities disabled, and these cards – which sell for $50 – reveal a code to download a software tool that unlocks those extra capabilities.

Not everyone approves. Some commenters (on more hysterical forums than ours) see this as an outrageous way to treat customers. “They’re selling deliberately crippled parts!” they cry. “They want to charge you twice for the same thing!” And, to an extent, they have a point.

A simpler, more flexible family

But for decades it’s been absolutely standard practice – and not just at Intel – to selectively disable CPU features, and restrict clock speeds, so as to sell the same basic chip at both the high and low ends of the market. All that’s new here is an officially-supported way to restore some of those features after purchase. If you think about it, it’s actually more empowering for the customer than the old way.

In fact, I’d love to see this approach applied across all processor families. Choosing a CPU need no longer be a leap of faith: you could start with a cheap model and add more power as you needed it.

It could simplify things too: instead of bamboozling customers with numerous poorly-distinguished CPU families, Intel could offer just three designs – say, an Atom, an LGA 1156 part with up to four unlockable cores and an LGA 1366 part with all the top-end features waiting to be enabled.

Technical challenges

Of course, there are potential technical challenges to such an approach. In the olden days, some slower and less capable chips were sold not just as part of a market strategy, but because they weren’t fully functional (the Intel 80486SX comes to mind). Obviously this would be no good for a user-upgradeable CPU: every chip sold would have to be capable of running at full speed with all its features enabled.

Every chip sold would have to be capable of running at full speed with all its features enabled.

But in 2010 this shouldn’t be a show-stopper. At the Westmere launch last year, Intel engineers confirmed that modern designs enjoy a huge amount of headroom — and indeed we’ve seen more than one Core i3 system running stably at 4GHz and above.

As for hardware defects, it’s hard to gauge how much of an issue this might be, as Intel doesn’t publish failure rates; but let’s just say it’s a long time since the company had any visible difficulty producing fully working chips.

The size of a die

Another consideration is die size. If every chip in a family had to support the maximum available feature-set, they’d all have to be built on comparatively large slices of silicon. The most lightweight LGA 1156 part, for example – the Core i3-530 – is etched on an 81mm² die. To match the features of the flagship Core i7-880, it would have to grow to 296mm² — more than a threefold increase in area. And, as chip manufacturers keep reminding us, a bigger die means a higher price.

But the silicon itself is nowhere near as expensive as you might imagine. It’s impossible to put a firm price on a square millimetre of wafer, especially since Intel does its own chemical processing; but in 2006 the website EE Times reported that 300mm silicon wafers – now the standard, but at that time still an upcoming format – were selling for $200.

If we assume (conservatively) that the price has remained the same, that means 81mm² of CPU-grade silicon would cost very roughly 57¢, while a 296mm² die would come in somewhere around the two-dollar mark.

Yes, it’s a price rise, but hardly one large enough to wreck the business. And the machining costs (which do represent a substantial investment) would be lower, because there’d be no need to set up a separate dual-core design. Intel could see a market benefit too, from offering a more flexible line-up, and enjoy a new revenue stream from customers unlocking processor features.

Catching on

For all these reasons, I dare to hope that the Pentium scratch-card represents a toe in the water, and that the approach will soon be extended to more aspirational processors.

If I’m honest, though, the possibility I’m most interested in is, from Intel’s perspective, the one that could kill the idea stone dead. Right now it’s physically impossible to turn a lowly Core i3 into a powerful Core i7; but in the model I’m imagining it would be a simple matter of entering a code. And, as we’ve seen with CSS, AACS and most recently HDCP, where there’s a code, there’s a concerted effort to crack it.

After all, you can already download dodgy code generators to unlock features in software. How many of us would scrupulously resist downloading a code for a free processor upgrade? And what on earth would that do to the CPU business?

As I entered my details this year I was pleased to notice that that particular data leak has been plugged. The information fields no longer support autocomplete, making it impossible to browse previously-entered information. A step up, surely, in security.

Sadly, the process remains fantastically insecure for other reasons. The computers themselves are regular Windows 7 laptops with full internet access – and they’re positioned facing away from the registration staff. So while it may no longer be possible to get information out of them directly, it’s a breeze to to download and install any software you wish (such as a keylogger or database scraper) without anyone being any the wiser. Indeed, since the registration machines are in a public lobby, you can just walk in off the street and start tampering with them.

Is this a serious problem? Admittedly, there’s probably a limit to how much havoc you can wreak on an IDF registration laptop. But if someone were to install an aggressive worm on one of these machines, it could easily spread to more sensitive systems, especially if the hardware ends up going back to Intel HQ. Good security practice means anticipating and eliminating risks like that.

Fundamentally, the lesson is the same as last time: when you build on an existing platform, your system inherits all the complexities and vulnerabilities of that platform. And, once again, this demonstration of that fact comes, poetically enough, from one of the pre-eminent producers of platform technology.

Here – with apologies for the ropey photography – are some of my favourites…

Now, before you imagine that IDF has been overrun by vandals, let me explain that Intel has installed whiteboards all around the conference centre, and has invited attendees to use them to share their visions of the future. It’s on these boards that the graffiti have been appearing. Some visitors have obligingly written up their hopes for new technology: a few arty types have even provided illustrations.But amid these contributions, there’s also been a spate of comments, on boards all over the centre, simply demanding: “Where’s Pat?” It’s a bit of mischief, but it also makes a serious point about corporate politics.

Predictably, Intel’s stewards have been erasing such scrawls wherever they’ve found them (adding to the whiff of Stalinist purge surrounding his departure). But my hat is off to one enterprising scribbler who correctly realised that he could sneak his protest past the censors by simply writing it in the international language of engineering…

Greetings from San Francisco! Back in the UK, I know most of you are probably gearing up to go home for the day; but out here it’s 8.15 in the morning and the Intel Developer Forum starts in 45 minutes. Over the next three days we’ll be learning more about 32nm CPUs, scoping out the successor (already) to Nehalem and – inevitably – enjoying more talk about Larrabee, Intel’s multi-core x86-based graphic system, now coming up to a glorious three years of development with no release in sight.

Still, leaving that aside, Intel’s doing pretty well right now. With Lynnfield barely out of the traps and a die-shrink already rumoured before Christmas, it’s clear that the company is, right now, at the top of its game in terms of innovation and engineering.

Which is why I was amused this morning when my very first interaction with IDF – the registration process – exposed a glaring security slip.

Checking out the check-in

On the face of it, it looked like a good demonstration of “the power of connected computing”, or some similar buzzwordy phrase. The company had decided to use a web-based registration system, and had lined up little kiosk computers, running IE8 and connected to the main server, in the entrance hall for attendees to enter their details.

The trouble was that Intel hadn’t disabled Internet Explorer’s “autocomplete” feature… so when I pressed “G”, to enter my surname, I was surprised to be presented with a lengthy drop-down box showing all the previous surnames that had been entered. It quickly became clear that, if I had the patience – and didn’t mind looking a little suspicious – I could easily use the information stored across the various kiosks to compile at least a partial list of people who’d already registered this morning.

All right, there’s probably a limit to how much mischief I could cause with such a list. But I guarantee you that if you were to ask Intel for this information it would refuse to provide it. Meanwhile, the technology it has chosen is leaking information, and compromising attendees’ privacy, in a wholly unmanaged way.

The moral, if there is one, is probably to be very wary of using general-purpose, off-the-shelf solutions for specific applications – because user-friendly features that are helpful in one context could be disastrous in others.

And since the world’s biggest provider of general-purpose solutions is probably Intel, I think this is quite a poetic start to the proceedings.