Benchmarking a Quad-Core Beast

During the last Intel Developer Forum in mid-March, I (admittedly, with a bit of envy) reported to you that Intel had invited a few select journalists to privately benchmark its then-unreleased Core 2 'Conroe' desktop CPU against AMD's then-premium processor. Although a few minor snafus got cleaned up later, the final analysis showed that Intel was on the up-and-up; the AMD system wasn't unfairly handicapped, the Intel system was comparable to what'd later ramp into production, and Intel had a winner on its hands.

Intel broadened the review pool a bit this IDF, and I was one of the lucky few folks selected to sit down for an hour in front of a pre-production quad-core Core 2 Extreme QX6700 system yesterday afternoon. This time it was an Intel-vs-Intel shootout; the Core 2 Quad versus an 'Extreme' dual-core counterpart. And, it's important to note upfront, I wasn't able to install my own benchmarks; the system ran a pre-configured BIOS, Windows XP build and application suite. However, despite an exhaustive perusal of the hardware and software configuration, I was unable to uncover anything shady, nor were any of my peers in the room at the time. And I suspect Intel's smart enough to know that if they tried to pull any tricks, at least one of us wouldn't be fooled….and the damning evidence would quickly be all over cyberspace.

Before continuing, you'll find snapshots of the Core 2 Quad system guts here and here. With the exception of a CPU swap, it looks identical to its Core 2 Extreme competitor. Here are the two systems' specifications:

Note that while the monolithic dual-core and dual-die quad-core CPUs run at identical 1066 MHz front-side bus speeds, the core clock speed of the Core 2 Quad processor is lower than that of its Core 2 Extreme counterpart. I was unable to get a straight answer from Intel as to why this was the case; I suspect that a desire to put a damper on the quad-core CPU's thermal design point was a motivating factor, along with a desire to maximize the dual-die configuration's composite test yield. However, unlike with Intel's prior, NetBurst-based dual-die offerings, a doubling up of the die within the Kentsfield package didn't lead to an added-impedance-induced decrease in the FSB speed. Intel proudly stated that unlike last time, Core 2 was designed from the beginning to be amenable to multi-die configurations. Judging from the 1333 MHz FSB of Woodcrest, the server-tuned spin of Core 2 now known as the Xeon 5100 series, Intel clearly also has more fundamental FSB headroom this time around.

Continued with 'Benchmarking: Titillating Results'….