The curtain has come down on Intel's first developer forum in Beijing, and it's been a blast. The trouble is that I've now got a hard disk full of notes, pictures and documents , but with no time to put the whole thing into context. That'll be next week's job – and like all good full-on experiences, the best time to write about it is in that golden window between catching your breath and having normal life take over again.
Here are some questions – and some entirely random answers.
Penryn – Intel's next big processor upgrade. I've seen demonstrations of working silicon in real systems, I've sat through the briefings and I've talked to the engineers. It's all dual core or quad core, but are there really no single core devices? With the emphasis on low power, surely a single core 45nm processor would find a healthy market in notebooks, where dual-core is really over-egging the pudding.
Nehalem – Penryn's successor. It's been around for a while (10GHz? Those were the days), but while it had a respectable number of mentions in the pre-show documentation and appeared on the schedule as a subject to be discussed, it was the ghost at the banquet. Not a sniff of info. Is it well? Have there been problems in its long gestation?
The Ultra Mobile Platform strategy. Having chewed on today's revelations for a few hours, I'm more confused than ever. The 2007 platform is based on a sawn-off Dothan – 90nm, for heaven's sake, two generations behind, with a quarter of the cache of the original part. Intel says that it's purely a transitional step, the last of the old order: it's good enough to snatch some wins from people like Via and keep some of Intel's pals on-side, but nothing more. Even Intel isn't using it in its Classmate PC.
The 2008 platform, Menlow, is shrouded in mystery. That's a brand-new design in 45nm, rebuilt from the ground up to give PC performance with substantial power and size advantages – yet it's compatible with the two year old Meron architecture, not Penryn, while still not slated to appear until the year after Penryn. That may not matter for the intended market for Menlow, which is "PC compatible" pocket devices capable of running Vista or Linux or (ahem) OS X, but it still leaves that curious single-core gap in the Penryn range. If you were thinking of making a decent sub-mini notebook with long battery life, what would you base it around? Not Menlow: Intel isn't even hinting that's a good idea, despite the fact that people are gagging for that sort of functionality. Instead, the company is concentrating its public Menlow push on the UMPC, a market which is tiny, badly defined and littered with corpses. How odd.
Many Core. Oh, where to begin? Everywhere you look in high performance computing, the future is massively parallel x86 (or as we must learn to call it, IA). It's a market due to grow like a teenager with a pizza addiction: it's absorbing jobs previously given to traditional mainframes as well as tasks that have out-grown their PC server or workstation childhood. Larrabee, Intel's as yet shadowy many-core IA chip, has been given a 2008 slot for detailed discussion: that means it's been in the works for a while. It's the stake Pat Gelsinger needs to drive through the heart of the stumbling zombie of Itanium – and the reasons this will happen are as numberless as the stars at night. If Itanium survives at all it will be like the orphan Xscale; palmed off on a company with an interest in its future, to placate partners and fulfil obligations. I had a bit of a barney with Pat over this – I said that Larrabee was a high performance computing chip, and he disagreed. It really is HPC through and through, and most of the applications Intel quoted for it are slap-bang in the middle of the HPC revolution, but the moment the company admits it... well, you can fill in the blanks.
But forget about Itanium (you have already, you scamp). What's particularly interesting about Larrabee – which I think is by far the most exciting thing to come out of IDF Spring 2007 – is the way the company illustrated it in the keynote. This is a chip capable of being a supercomputer, but cheap enough to go into the home. What to do with all that power? It was shown as a bank of thirty six IA cores dedicated to watching football and automatically preparing edited highlights. Two-thirds of the cores were dedicated to identifying players and the ball itself. Fair enough. A couple of cores were running security and virus scanning – again fair enough, given the sort of nonsense that goes on. There were a couple encoding video, and four enhancing the picture.
But another four were marked down for 'user interface'. Now, four IA cores is a heck of a lot of welly, especially when employed soley to work out what to tell the user and how to respond to their commands. That's the sort of thing a 4MHz eight-bit micro did perfectly well back in the day, when all that was required was watching the keyboard and writing a few words to the screen. So these four cores are doing something else. Speech and video recognition is my guess: watching how you move and gesticulate, and responding to your words of command. That's some video recorder.
This is technology due to be demonstrated next year. Mmm.
I could be over-analysing. But in my experience, Intel is very careful not to say things in keynotes that it isn't really thinking, although it's perfectly happy not to draw attention to the details.
Other things that are still worth mulling over: Intel's addiction to WiMax, and admission that it's going to be a tough call in Europe – and is Estonia really at 85 percent WiMax coverage already, as I was told tonight? How come everyone on stage at this IDF wore suits and ties, when previously it was open-neck shirts and an interesting variety of trouser? (OK, I know the answer to this – but the cultural side of an IDF in Beijing is worth an entire essay in itself).
And why am I up writing this at quarter to four in the morning, when Air China is expecting my presence in a mere ten hours' time?