enigma067 :
I wouldn't compare a Multi-Chip Module (Q6600) with a Native Quad Core (Phenom) if I were you.
The Phenom's architecture is far superior to anything Intel has out now.
The AMD K10 μarchitecture is on par with Intel's Core μarch. The execution cores have many of the same features and, surprise surprise, they perform pretty similarly clock-for-clock, give or take a few percent depending on the particular code you're running and the amount of cache bolted to the core.
Intel is good at two things; making more expensive chips and regurgitating the 8086 architecture.
First of all, the new ISA standard is not Intel's but AMD's x86_64. Intel had wanted to go from x86 → Itanium IA64 but AMD captured the market with their x86_64 ISA. I'm not going to argue with your second point as Intel does on average charge more for their chips than AMD does and usually has since they started making CPUs. But you did forget the thing Intel is actually REALLY REALLY GOOD at- IC photolithography process engineering. Intel is world-class in their fab and process technology. Maybe IBM can go toe-to-toe with them, but Intel beats IBM in execution of the latest fab technology to make hundreds of millions of ICs.
I can't wait to see how expensive their 45nm chip is going to be.
Probably about as expensive as equivalently-clocked Phenom X3s/X4s and Opteron x3xx CPUs, maybe a tad less.
Some people overclocked the Barcelona to 4.7Ghz under liquid cooled environment.
Barcelona like Phenom is Native Quad Core.
And people got a Skulltrail with dual QX9775s to 6 GHz under a liquid-cooled environment (that liquid was nitrogen, but it's a liquid, right?
) It really makes little difference to anybody who actually uses their computer how high somebody can get a CPU to go in a suicide shot using a grand worth of LN2 and a Vcore approaching two volts. Really the only thing that matters would be how high somebody can clock the chip on a sane Vcore using sane cooling methods, which usually means a large air-cooled heatsink or maybe water.
AMD will tell you themselves that MCMs are not bad as they are readying a 45 nm 12-core Opteron made of two 6-core dies in an MCM. The IMC might make some performance difference and the Opteron's HT bus setup vs. the Xeon's FSB setup does as well, but you can implement either using a native multi-core die or MCMs. There are advantages to a multi-core monolithic die, don't get me wrong, but they are not the end-all be-all in chip design. The implementation of buses, core-to-core communication, caches, core μarch, and other features of the chip make more of a difference than if there are two dies or one under the heat spreader.
Just_An_Engineer :
I have a couple of comments myself. You are correct that Intel is good at making expensive chips and reusing the x86 architecture. However, the chips Intel has been making lately are very good even if they are more expensive. Everyone, including AMD, still uses the x86 architecture for desktop CPU's.
Almost everybody is using x86_64, which is a derivative of x86 but has considerable differences.
Also, x86 is an instruction set, not an architecture. An 80386DX has a *wildly* different architecture than a Core Duo but both are x86 processors. The instructions were separated from the hardware a long time ago when decoders were introduced in the first P6 chip, the 1995-era Pentium Pro. Today's "x86" cores are really much more like RISC cores rather than CISC ones like the x86 instruction set would lead you to believe.
In my opinion the native quad core design is superior to the multi-chip module that Intel currently uses. The problem is that while AMD's concept for the design was great, their execution was disappointing. Intel will be using the same conceptual design in their next processor generation and it remains to be seen if they will fare any better.[/quotems]
It will be interesting to see the Nehalem launch. It's a big change for Intel as far as a macro-architectural and platform standpoint and there are supposed to be μarch tweaks as well.