System setup and notes
Hardware
- Corsair 1GByte DDR2 Twin2X Matched Memory Pair, running at both DDR400 and DDR533 speeds
- Intel Pentium 4 560 - 3.6GHz
- NVIDIA GeForce 6800 GT 256MB, PCI-E, 350/1000
- ASUS P5GD2 Premium i915P LGA775 motherboard. run in non-Hyper Path 2 mode
- Samcheer 420w PSU
- LGA775 reference cooler
- WD 160JB 160GB hard drive (PATA)
- Dell P991 19" flat-faced CRT monitor
Software
- Windows XP Professional w/SP1
- Intel 6.0.1.1006 chipset drivers
- NVIDIA ForceWare 61.76
- DirectX 9.0b Runtime
- Pifast calculation to 10m places
- ScienceMark 2.0
- 3DMark2001SE
- Unreal Tournament 2003 Retail (patched up to 2225 - HEXUS custom benchmark)
- Quake III
Notes
It's difficult to know just how to benchmark a new memory standard. Comparing performance with regular DDR memory is problematic because the test motherboard, an ASUS i915P Grantsdale, runs solely with DDR2 memory. Comparing it with, say an AMD Athlon S939 CPU and low-latency DDR memory skews the issue by introducing platform-specific differences. Therefore I've decided to see just what kind of gains DDR2 memory running at an effective 533MHz has over identical memory at 400MHz.Theory suggests that differences should be minimal, for Pentium 4 560s, which use a 200MHz FSB, run optimally with synchronous DDR400 RAM. Adding in more memory bandwidth via an uprated memory controller adds to clock buffering and overall latency. The question is whether more bandwidth can offset the negative aspects arising from asynchronous FSB and RAM speeds.
In summary, then, Corsair's DDR2 pack was run in the following combinations:
3600MHz CPU - 200MHz FSB - Corsair DDR2 Twin2X @ DDR400 (4-4-4-12)
3600MHz CPU - 200MHz FSB - Corsair DDR2 Twin2X @ DDR533 (4-4-4-12)
Dual-channel memory running at an effective 400MHz gives a potential 6.4GB/s of bandwidth; identical to the processor's maximum uptake.
An effective 533MHz affords around 8.53GB/s of bandwidth. Is more always better?. Let's find out.