Linked by Thom Holwerda on Wed 18th Jan 2006 19:00 UTC
Intel The VP of Intel's digital enterprises group told the INQ this morning that the Itanium microprocessor is starting to make waves in the corporate enterprise market. Kirk Skaugen, of the servers platform group, showed a slide which claimed the Itanium processor was eating into Sun and IBM Power shares, based on "customer revenues". He also said that since the fourth quarter of 2003, applications for the Itanium family had grown to 5900 by the end of 2005.
Thread beginning with comment 87366
To view parent comment, click here.
To read all comments associated with this story, please click here.
RE[2]: Uh huh
by nimble on Thu 19th Jan 2006 07:14 UTC in reply to "RE: Uh huh"
nimble
Member since:
2005-07-06

Nobody, save perhaps for AMD*, foresaw the clock-rate wall that the industry would hit. Who would have expected getting only a 2.5x scaling from four process shrinks, when previously a 3x-4x scaling had been had from only two?

To be precise, it wasn't the clock-rate wall as such, but the power wall. Clock rates used to be limited by transistor switching speeds, but with the P4 they're limited by power consumption. The P4 could probably run at 6GHz right now, but the cooling for that would just be too expensive and loud.

Intel must have foreseen the increase in power consumption, but they probably expected process shrinks to make up for that. But that hasn't happened.

The unfulfilled 3GHz predictions and the liquid cooling suggest that the G5 with its 16-stage pipeline has hit the power wall to.

AMD's processors with their 12-stage pipeline don't have such big power problems, which means they're probably still hitting the switching speed wall first.

I guess for maximum performance you'd want to hit both walls at the same time.

Reply Parent Score: 2

RE[3]: Uh huh
by Phillip.Fayers on Thu 19th Jan 2006 10:59 in reply to "RE[2]: Uh huh"
Phillip.Fayers Member since:
2005-12-14

Intel must have foreseen the increase in power consumption, but they probably expected process shrinks to make up for that. But that hasn't happened.

Quote from a New Scientist article titled Has Moore's Law had its Chips? (vol 182, issue 2448, p26, 22nd May 2004).
<blockquote>"Somewhere between 130 nanometers and 90 nanometers the whole system fell apart. Scaling stopped working and nobody seemed to notice." - Bernie Meyerson, Chief Technologist, IBM.</blockquote>
People did predict the power increase, but they didn't predict all the problems caused by the process shrink. Have a look at a paper from ISSCC 2001 Microprocessors for the New Millenium: Challenges, Opportunites, and New Frontiers:
http://ieeexplore.ieee.org/iel5/7284/19686/00912412.pdf

At one point Gelsinger writes about power consumption:
<blockquote>... predicted powers are excessive and prohibitiveley large for any practical application, and it is clear total power consumption will become a limiting factor.</blockquote>
But he concludes the paper with:
<blockquote>if is boldly predicted that the microprocessor of the year 2010 will have 1B transistors on a die, operate at 20 to 30GHz, and perform over 1T operations per second.</blockquote>
Even though everyone expected problems they had 20 years of experience of solving such problems and were confident that they could solve this one.

Reply Parent Score: 2