Intel chief: 64-bit could take ten years to conquer desktop

Gordon Moore is alive and well -- and so is his law, Intel boss Craig Barrett said at the Gartner ITExpo in Orlando yesterday, dismissing the recent suggestion by an Intel engineer that Moore's time was up. In fact Barrett envisaged five more generations before the limits of CMOS technology were reached, when the signal to noise ratio would become critical at around 0.05 microns. By most standard this is pretty optimistic, and provided Barrett isn't just whistling in the dark it suggests that Intel must have several rabbits it thinks it can pull out of its R&D hat. But as a consequence, Barrett thinks that the world will come to an end on 15 June 2015 - or at least CMOS will. For Intel, GHz chips were something for the second half of next year, Barrett said, although he liked to refer to these as 0.001 TeraHz processors. Asked about Itanium, he mentioned that as a metallurgist, he worked quite a bit with Titanium, which seemed to be a clue. Certainly he preferred the name Itanium to Celeron. He did have some ominous words to say about the move to 64-bit, suggesting that it could be five to ten years before it would be the norm on the desktop; he compared this to the ten years between the introduction of the 80386 and the significant development of 32-bit desktop applications. If it's really going to take ten years, then we can maybe start wondering about Intel's roadmapping. The company officially reckons McKinley, the iteration after Merced, will be the 64-bit Intel chip that moves into the mainstream, but that's hardly a ten years from now product. Intel does intend to keep plugging away at 32-bit alongside 64-bit for a few years, so Barrett could be signalling that this plan will be extended. Or alternatively, we could bear in mind that Intel was happily selling 386 chips for several years before the software started to catch up. But there were reasons for the slow development of the software that probably don't apply these days. At the 386 introduction Microsoft and IBM were in the driving seat (and incidentally still squabble over whose fault it was they originally aimed OS/2 at the 286). These days we've got Linux and several flavours of Unix aimed at IA-64, so take-up is likely to be a lot faster. But that's software, and a problem for Microsoft - so Barrett should worry. On bandwidth his line is down-the-line Intel. It will not only increase, he forecasts, but will become essentially free. Of course, the advantage will be that more powerful processors will be needed to compress, decompress and process the data. He also observed that France had at last emerged from the 15th century so far as encryption was concerned. Unfortunately, he was very gently questioned, so when he was asked about the Intel-Microsoft relationship, and whether Microsoft was becoming less Intel-focussed, he only genuflected and swore Intel would not develop an operating system. Nor would Barrett elaborate on whether Intel was taking any design steps to assist Unix flavours. When asked about Linux, he trotted out the usual bullet-point attributes, but Linux has some way to go in the processor-design space. Barrett said that Intel was one of the major users of workstations in the world, and had evolved from IBM to HP and now to NT machines, but with some Linux. He said he had a loose marriage with software. A question begging to be asked that was unfortunately not pursued - that e-commerce would demand great scalability in servers, and was NT up to it - might well have elicited an interesting reply. ®