Intel to delay next-generation mobile chip

TalkBack 4 of 4:: « Previous

Thread View
Flat View

Gotta wonder: has Intel hit the wall?: There has been no significant increase in the speed of Intel's chips since the 3GHz Pentium 4 that was released about 12 months ago. They've released Xeon and Itanium chips in the meantime, but they aren't faster, though they are a bit more powerful for server operations.

Moore's law does not actually apply to speed (MHz) or power, but to the number of features (transistors) on a chip. It just happens that the number of features is closely correlated to performance. Originally (in 1965), Moore said they'd double every 2 years, then revised it to every 18 months. Chips have, more or less, kept pace ever since - Intel says they can keep it up until 2010.

The Pentium 4 of 2000 had 42 million transistors, to keep up with Moore's Law, Intel should have been at 84 million in mid-2001, then 168 million by the end of 2003. By the middle of 2005 they should be well over 330 million and at 2.7 billion by 2010 using a 45nm (that's 45 nanometre or .000000045 metre or 45 billionths of a metre) process.

Associated with each increase in the number of transistors is a decrease in the minimum feature size - these steps happen every 2 to 3 years. Intel have less than 6 years to get from 130nm through 90nm and 60nm to 45nm - they seem to be really struggling with thee step to 90nm.
Intel achieved 130nm in 2000 and should have had 90nm 2003 but still haven't done it in early 2004. If this continues, 30nm will arrive in 2020, about 8 years late. Clearly some other tactic is required to increase the power of computers, since the mechanical approach of increasing transistors and reducing feature size seems to have hit the wall.

This focus is already showing signs of weakness - Hyper-threading is sorely needed in the Pentium architecture because of the very deep (20 stage) pipeline, compared to Athlon and PowerPC's 7 stages. In the 64 bit world, PowerPC G5 and Opteron deliver mixed 32 and 64 bit results superior to Itanium's pure 64 bit efforts - Intel needs to pull their socks up.

Perhaps we need a much stronger focus on multi-processor systems and wider buses? Maybe we'd be better off aiming for 4-way or even 16-way cores with 128 bit data paths? If so, I'd bet on IBM taking over Intel's dominance of PC chips by 2010.

Maybe IBM should enter into an agreement with AMD where IBM do the underlying grunt work and AMD do the front-end i386 compatability.; Posted by: Fred Fredrickson Posted on: 01/15/04 You are currently: a Guest | Members login | Terms of Use