Last day, and everyone's been Intelled out. It's my favourite day, though, because it's the R&D keynote with technical director Pat Gelsinger. Hairstyle reassuringly unchanged, he leads the faithful through yet another of his Big Visions. We are entering, he says, the Era of Tera. That slightly uncomfortable phrase -- it looks like it should rhyme, but Tera is pronounced to rhyme with cellar, not clearer, so it ends up sounding like the Era of Terror -- is his shorthand for terabytes, teraflops and terabits per second. He leads us through the three previous ages of Homo Siliconia -- the Kilo, where everything was text; Mega, when Windows popped up; and Giga, where we are now and where multimedia is king. I realise with something of a lurch that I've been there for all of them -- my early days in IT are now consigned to the rough equivalent of the Neolithic. Thanks, Pat.
It's all classic stuff. We can't make the processors much faster, they'll melt and the memory won't keep up -- which everyone knew, but it's quite something to hear it from the top in such an uncompromising fashion. Not 'we can keep squeezing stuff out', but 'that's it. We're hitting a wall'. Of course, he has an answer -- massively parallel multicore chips, supersmart cache mechanisms, photonic interconnects -- and even an example of what we'll do with all this power. On comes a bearded European who shows off real-time raytracing, where a couple of Volkswagen Beetles have been dumped in a medieval Florentine chapel and made to look very shiny indeed. It's one of those moments where you realise that what you thought were really nifty graphics in your computer games were in fact miles behind what you want -- and you want this.
But this requires -- and here the good doctor unveiled a huge rack of computers with a flourish -- 400 gigaflops of processor. He'd really like a teraflop, if it's all the same to you. And then we can play Quake on it.
The fun bit is that Pat has to systematically diss -- his word -- most of the current technologies propping up the Intel House O' Fun to make way for the bold new world. He takes one particular example, TCP/IP Offload Engines, or TOEs. These are bits of silicon that take over network packet management from the main processor, and were being heavily promoted as the best way to speed things up a while back. They didn't work very well, and Pat proceeded to demonstrate why -- the problem they fixed, processing the packets, wasn't the problem that was really there, which was moving the packets around the place. I momentarily dream of a headline, Pat Snubs TOE. At the end I ask Gelsinger why, if TOEs didn't work, they'd spent so long promoting them -- well, we've only really just done the analysis came the question-begging reply.
As I walk out of the keynote in the company of Nick, a chap in an AMD sports shirt comes up to me. "That stuff about the TOE was done ten years ago by another company, you know," he said, to the visible bristling of my PR pal -- whose sense of humour had been thoroughly tested by the AMD onslaught. Well, yes. I did know -- a long time ago I'd been involved in the design of network chips and software -- and freed from the undeniable charisma of Gelsinger's stage presence I realised that not very much new had been said. It had been very competently presented as a coherent thesis, and was all worth saying, but it left a great deal unanswered
If Intel is going to embark on a radically new direction where parallel processing is king, how will it make the transition? It's had enough trouble going from 32 to 64 bits, which is rather like learning to drive a bus after mastering a VW Beetle. Going to the parallel world will be like learning to fly a 747. But there really is nowhere else to go. The other thing I realised was that this had been the only keynote where nobody had asked about 64-bit Pentium extensions afterwards. Ya gotta love that Gelsinger magic.
Still mulling over these thoughts, I make my way to the final session of the IDF. This was all about wireless -- and the very prosaic business of testing 802.11. Intel has gone about this methodically, creating three test areas -- an office in a city, a radio-screened room and a field in the middle of nowhere -- and set about measuring everything that can be measured while the network does all the sorts of things that wireless networks do. The office is laid out like a typical American workplace with chairs, desks, partitions, walls and equipment all where it should be, but without any people.
A German voice piped up from the back of the room. "Do not people affect the radio?" Yes, said the engineer, but not very much. We've done tests and there isn't much difference with or without, so we're doing it without. "But that is not typical?" said the voice. Good point, said the engineer, and one day I'm sure we'll factor that in, but there are more important things to do first -- this in the "let's move on" tone of voice you get to recognise very quickly towards the end of the show. Our inquisitive Teuton was having none of it . "This is an office, so there must be people. And at two point four gigahertz, they will be affecting the radio." It will come, said the exasperated engineeer, whose fondness for rooms devoid of human life was clearly growing by the second. Now, if we look at this graph…
"You could use sacks of water. Possibly with salt in." said the voice. I didn't catch exactly what the engineer then said, but there was a sound not unlike that of a sack of water -- possibly with salt in -- being dragged outside.
We moved on.
As must I. I wish there was time to tell you about the evening in the Japanese restaurant where a fresh-faced young UK journalist was introduced to sashimi for the first time ("But it's RAW!"); the ongoing battle between Them Wot Smoke and Them Wot Won't Let Them; and the wake for our dear departed friend Lynne, which covered three continents linked by mobile phones and ended up with one of our number discovering the hotel sold gin and tonic by the pint. Repeatedly. Then there was the laptop with the pink satin curtains, the briefing rooms that looked exactly like the bogs at Glastonbury, and the automatic paper towel dispenser in the real loos that proudly declared itself "Motion Activated". Now there's automation for you. And if only I could repeat the conversation between a very keen British hack and a very cute American woman, that started off as a very promising chance of a pull before it drifted into politics -- yes, Iain, lots of these people really do like Dubya -- but I'm being thrown out of my hotel room in ten minutes and I haven't even packed my iPod.
Another time, perhaps. See you in Blighty.