Mark & John - Re: "hitting the wall of superscaler architechure..."
The article never mentioned the Pentium Pro. Indeed, it implied that ALL CURRENT CPU architectures were approaching the same limits, producing fewer instructions/per clock improvements based on architectural features alone.
The only processor mentioned, by example, was the DEC ALPHA chip, the text of which is listed below (extracted from EE TIMES/TECHWEB):
{=================================================================}
"But all of the complexity of superscalar processing and the techniques used to accelerate it appears
to be reaching diminishing returns. The average number of effective instructions per clock has not
gone up that much in recent chip announcements. One paper observed that running Digital
Equipment Corp.'s four-way superscalar, 300-MHz Alpha 21164 actually ran at two clocks per
instruction, not four instructions per clock on an SQL application. Equally serious, the design time for
new CPUs is skyrocketing. "
{====================================================================}
Ron Wilson, the author, suggested that real speed improvements have come more recently from process improvements, meaning - FASTER CPUS based on smaller geometries.
From this standpoint, INTEL is in the CATBIRD seat.
Intel doesn't own any MEGAFABS. They just have three or four billion dollar+ fabs cranking out 0.35 micron Pentium/Pentium Pro chips with a new fab scheduled to bring up 0.25 micron Pentium Pros next year.
Memory access speeds and other issues such as compiler efficiency were cited as more fruitful endeavors for faster performance. Here too, Intel is in a good position. Their innovative GTL+ memory interface, split bus architecture, and L2 cache within the CPU package on the Pentium Pro, although expensive, produces good performance gains.
As for architectural improvements, Intel is working on the P7 now and has a cadre of researchers assembled to develop architectures beyond the P7. And these researchers are motivated by the same issue as all of us - their stock options will make them rich if the company's sales soar due to their successes. GO RESEARCHERS!
I'd say that Intel is not sitting on its laurels, and when the x86 architecture is supplanted/replaced, Intel will be the one to replace it. They have a long history of eating their own children.
Paul |
