Jim - Re: "Do you really think that the Timna is something Intel should pursue? Look what happened to NSMs PCOAC. Vaporware and if and when it appears it will be in the "MHz loser" category and be relegated to the no profit low end. "
The short answer is YES !
Bear in mind, Intel is NO NSM. NSM had no experience in successful x86 manufacturing - and certainly NO EXPERIENCE in successful CPU design - their 32xxx series was buggy from birth to death. And that plague was inherited by Cyrix, it seems.
Intel originally rejected the "integrated" approach after some technical and marketing mishaps with their 486SL technology. At that time, many of Intel's customers were upset that they couldn't customize ("add value") to Intel SL designs (primarily for mobile applications) and concurrently, Intel was faced with much higher costs due to adding all the bus driver circuits to the 486.
When Cyrix introduced the MediaGX, Intel figured they (Cyrix) would meet the same fate - but they didn't.
Compaq - which had balked at Intel's earlier approach - sucked up the MediaGX. Of course, having Gary Stimac on the CYrix board (Stimac was a long time Compaq Engineer and VP) certainly "helped", as well as the fact that the Cyrix name did not contain the letters I N T E L .
Then, the expenses of the Pentium II packaging - with off-chip L2 cache - cut into Intel margins, at the same time as Intel's yields were heading for the stratosphere.
Somebody at Intel had an epiphany - that adding the L2 cache directly to the Pentium II could be done ECONOMICALLY with Intel's exceptional yields thereby LOWERING overall CPU costs - allowing Intel to offer a lower cost Pentium II/Cache solution for the "low end" - now known as Celeron.
[Note - earlier successes with 8087 and L1 cache integration onto the 80486 were helpful here.]
The first Mendocino was such a huge success - first silicon was essentially production worthy - and yields were still quite favorable, that Intel realized that the KEY to their LONG TERM SUCCESS was their MANUFACTURING TECHNOLOGY !
This MANUFACTURING TECHNOLOGY could be leveraged in several ways, and the further INTEGRATION was a way to do this.
The overall silicon costs increase with integrated die, but the overall SYSTEM COSTS DECREASE !
And Intel thereby becomes even more a part of the overall SYSTEM COMPONENT provider if they (Intel) do the SILICON SYSTEM Integration.
Whitney is a case in point. In the next quarter, Q3, i810/Whitney Motherboards will pop up so fast that Jerry Sander's Mercedes will go into a tailspin.
With that approach - CPUs with L2 cache and Chip Sets with Graphics, the natural extension is to combine them both into one big CHIP.
Intel's question to themselves was: "If not us, then who? "
The dangerous decision was to WAIT and find out who that "who" would be.
The SMART DECISION was to make sure that "WHO" was INTEL !
And it will be Intel.
The inevitable approach - assuming it is economically feasible - is to integrate as many PC functions as possible onto one piece of silicon. Intel saw the opportunity was theirs for the taking as they were sitting on the largest Silicon Manufacturing Machine in the world coupled with the, most likely, the world's HIGHEST YIELDS.
To ignore that opportunity would be foolish, and Intel has a history of betting on new products that will inevitably obsolete their old products. And almost every bet has been a winner. Witness the earlier successes with 8087 and L1 cache integrated onto the 80486 .
The RAMBUS interface now offers reduced "pin count", hence reduced bus drivers and lower silicon real estate to implement the memory interface.
And keep in mind - Intel's 0.18 micron process is being ramped up in FOUR fabs today.
The 0.13 micron process is in full blown development status.
And Intel will be targeting a new Fab in Oregon, D1C, for TWELVE INCH WAFERS using that 0.13 micron process. Hopefully, these will come to fruition in about 2 or 3 years.
Thus, Timna will prove out the concept of a nearly FULL PC on a chip and its follow-ons - 2 or 3 years down the road - will be made on 0.13 micron process on 12 inch wafers at VERY LOW COSTS.
That is the future.
And Intel is positioning itself to get to the future before anyone else.
Paul |
