Test Setup
Processor	Intel Pentium 4 2.0 GHz
Mainboard	DFI NB70-SC (i845D) ECS P4IBMS (i845)
Memory	256MB OCZ PC2700 DDR SDRAM 256MB Mushkin PC133 SDRAM
Video	nVidia GeForce 2 GTS 32MB
Hard Disk	IBM Deskstar 75GXP DTLA-307020
Other	Creative Labs SoundBlaster Live! Value SMC Etherpower 10/100 PCI NIC
Operating System	Microsoft Windows XP Professional

We’ll be using three platforms to conduct our scalability testing, one equipped with PC133 SDRAM, one with PC2100 (DDR266) DDR SDRAM, and a final platform with PC2700 (DDR333) DDR SDRAM. Intel will shortly be releasing higher speed grades of the Pentium 4 with a faster 533 MHz FSB, and faster DDR memory, so our intent here is to give some sort of an indication as to what to expect on that front.

In keeping with that, you’ll notice four ‘platforms’ on most of the graphics. Here’s a breakdown of each:

400/133 – Brookdale SDR (845) platform. FSB at 100 MHz (400 MHz effective), memory clock at 4/3 FSB, or 133 MHz.
400/266 – Brookdale DDR (845D) platform. FSB at 100 MHz (400 MHz effective), memory clock at 4/3 FSB, or 133 MHz (266 MHz DDR effective).
500/266 – Brookdale DDR (845D) platform. FSB at 125 MHz (500 MHz effective), memory clock at 1/1 FSB, or 125 MHz (250 MHz DDR effective).
500/333 – Brookdale DDR (845D) platform. FSB at 125 MHz (500 MHz effective), memory clock at 4/3 FSB, or 166 MHz (333 MHz DDR effective).

The latter two setups use a 500 MHz FSB. This is intended to give an approximation of upcoming processors that will use a 533 MHz FSB. We were forced to use 500 MHz, as opposed to a true 533 MHz, in order to keep the memory clocks within a more realistic range.

Each benchmark was conducted on each platform with processors using 14, 16, 18, and 20x multipliers (effectively 1.4, 1.6, 1.8, and 2.0 GHz) in order to map out scaling as clock speed is increased.

Each benchmark was executed three times, and the results averaged. Our test platform was set up using a Windows XP Professional installation. The only additional drivers installed were Intel’s INF Update, version 3.20.1008, and nVidia’s Detonator drivers, version 23.11.

Finally, many of you will note that the scales on the graphs do not start at 0. This was intentional on our part, as focusing in on a particular part of the scale helps to exaggerate the curvature of the lines, which, in this case, is more important than a comparison of absolute performance.

Thanks to OCZ Technology for providing the PC2700 (DDR333) memory used in testing.