Memory: Is more always better?

Test Setup:

AMD Athlon 64 FX-60+ (operating at 2600MHz, 13x200MHz); ASUS A8N32-SLI Deluxe (NVIDIA nForce4 SLI X16); Gainward PowerPack! Ultra/3500PCX Golden Sample (GeForce 7800 GTX operating at 470/1300); Western Digital Raptor 74GB, 10000RPM SATA 150 Hard disk drive; OCZ PowerStream 520W Power Supply; Windows XP Professional Service Pack 2; DirectX 9.0c; NVIDIA nForce4 SLI X16 standalone version 6.82 WHQL; NVIDIA beta Forceware version 82.12.

Memory used:

• 2x 1GB Corsair XMS4000 Pro operating at 400MHz 2.0-3-3-7 1T;
  
• 2x 1GB Corsair XMS4000 Pro operating at 520MHz 3.0-4-4-8 1T;
  
• 4x 512MB Corsair XMS3200XL Pro operating at 400MHz 2.0-2-2-6 2T;
  
• 2x 512MB Corsair XMS3200XL Pro operating at 400MHz 2.0-2-2-6 1T.

We're trying to find out whether there is a benefit from upgrading your existing 1GB configuration to 2GB using two 1GB modules, or whether you'll be able to get away with buying another two 512MB modules to run four DIMMs together using the 2T timing.

The 2T command rate timing is forced on Athlon 64 processors when all four memory banks are populated and can result in a performance drop over 1T, available when using just two memory banks.

We're going to attempt to measure real-world performance differences between the four memory configurations in Battlefield 2, Call of Duty 2, Day of Defeat: Source and F.E.A.R. to see if there is any noticeable difference. We overclocked the two 1GB modules to 260MHz (520MHz DDR with 3.0-4-4-8 1T) while keeping the CPU speed constant at 2600MHz in an attempt to see whether increased memory bus speeds has an affect on real-world gaming performance.

Unbuffered Memory Bandwidth:

We ran SiSoft's Sandra unbuffered memory bandwidth test to give us a yard stick for our gameplay evaluations. It was no surprise to see that the two 1GB modules at 520MHz outperformed everything else by a long way. The slightly tighter timings on the two Corsair 3200XL Pro modules meant that there was a slight bandwidth improvement over the two 1GB modules at 400MHz. However, this test doesn't take memory size into account - that's something that the games should do.

The four 512MB modules were the slowest of the bunch due to them being forced to run with a command rate of 2T. The drop in performance was around the 10% mark - we'll see whether this performance drop holds true across our selection of games.