Intel Prescott contains non-AMD 64-bit tech – report
Update Intel's upcoming (it hopes) 'Prescott' processor already has 64-bit extensions to the 32-bit x86 instruction set, but they're (a) not enabled and (b) not based on AMD64.
So says Xbit Labs, citing an anonymous source.
We've heard such claims before, usually in emails from folks unwilling to supply their name or give any indication about how they might know the information to be true.
Intel's official line is that 64-bit computing isn't needed on the desktop today, and won't be for a few years yet. Both Apple and AMD disagree, but it's worth bearing in mind that Apple's latest operating system, the rather good Mac OS X Panther, isn't a full 64-bit OS; it merely has enough 64-bit-ness to allow the G5 processor to access more than 4GB of memory. As for AMD's 64-bit OS, well, Microsoft is hardly rushing to get 64-bit Windows XP out the door.
In short, neither Microsoft nor Apple are wholeheartedly embracing 64-bit as the future of the desktop. Instead, they are understandably playing cautious, testing the market to see what demand there is (in Microsoft's case) and leveraging what the technology can bring to pro customers (in Apple's). Only AMD is tub-thumping for 64-bit, and that's because its future success depends on it - although it always has the 32-bit compatibility 'Get Out of Jail Free Card' if things don't play out as anticipated.
The bottom line then is that Intel is probably better off waiting, like Microsoft, to see how demand for 64-bit on the desktop pans out. Yes, it has a non-x86 64-bit platform already, but Intel is enough of a spinmeister to know it can easily get away with shipping a second, x86-based 64-bit system providing it pitches it right - i.e. at desktops, workstations and low-end servers.
It's never been inevitable that 64-bit would come to the desktop, but it's a logical extension of the evolution of desktop computing we've seen to date, first 8-bit, then 16-bit and now 32-bit. Intel can see this, and will have a desktop 64-bit contingency plan in place. It also knows the importance of the x86 software base in the commodity end of the market, and understands the importance of backward compatibility.
So some sort of 64-bit extension to the x86 ISA is its logical course of action. It builds on its dominance of the 32-bit space, but still leaves the company plenty of room to market Itanium as a high-end solution. Assuming it doesn't ditch the EPIC architecture, of course - but that's another story.
Enter 'Yamhill', the long-rumoured and much speculated upon Intel answer to AMD64. It almost certainly exists, but it's not known if this is anything more than a research project. The comments from Xbit Labs' source - assuming he or she knows what they're talking about - suggest that it has moved out of the lab into the fab.
Not so long ago, American Technology Research analyst Rick Whittingtonsaid Intel would implement Yamhill in 2005. That's the timeframe for 'Tejas', Prescott's successor, and also around the time Intel expects applications to start requiring more than 4GB of RAM - ie. "several more years from now", as Intel CTO Pat Gelsinger recently put it.
Intel's upcoming 775-pin Land Grid Array pin-out, to be implemented in the upcoming chipset 'Grantsdale', perhaps suggests a wider address bus. The Athlon 64 by contrast has 754-pins, while the current Pentium 4 uses just 478. Grantsdale is being primed for Tejas, but there's a Prescott version coming too. So if Tejas has 64-bit support, so it would appear does Prescott. The shift to 90nm should give Intel plenty of space for 64-bit support, particularly if AMD can do it at 130nm. The extra transistor budget can't have been spent exclusively on extra cache and Prescott New Instructions, surely?
The remaining question is whether Prescott and Tejas use Yamhill or AMD64. Some commentators have claimed Microsoft will only support one 64-bit x86 platform, but equally it's hard to see Intel being happy choosing someone else's technology - and it's less likely to give such a public thumbs-up to its rival's platform. 'What, you mean we could have bought Athlon 64s all along?' punters will ask.
Microsoft, meanwhile, isn't likely to be willing to pass on selling 64-bit upgrades to all those Prescott and Tejas owners, so we shouldn't put too much faith in an Windows-AMD64 axis.
AMD fanboys can spin all this in whatever way they care too, while they moan about Intel's 'malevolence' or Apple's 'lies'. The company itself has a rather better grip on reality and knows it needs to maximise whatever time Intel allows it to build up its 64-bit credentials and user base.
Some folk will complain about Prescott's disabled 64-bit support, but AMD is expected to do just the same when it ships Hammer-based Athlon XPs next year. It's not cheating, it's maximising your output, and so long as neither company markets their chips as offering more functionality than they do, then that's OK. It's good business.
After posting the above, we see (via Linuxworld) Lehman Bros. analyst Dan Niles is predicting that Intel will release a 64-bit x86 part sometime during the first half of 2004 - well ahead of the 2005 timeframe. He too reckons it won't use AMD64. ®