Test Setup

We attempted to keep a consistent CPU speed through testing, however due to limitations in chipset straps between CPU front side bus and memory bus when overclocking, a closest fit was applied where necessary.

Memory speeds tested:

• PC2-800MHz (low latencies) 2.93GHz/266MHz FSB;
  
• PC2-1066MHz (low latencies) 2.93GHz/11x266MHz FSB;
  
• PC3-800MHz (low latencies) 2.93GHz/11x266MHz FSB;
  
• PC3-1066MHz (low latencies) 2.93GHz/11x266MHz FSB;
  
• PC3-1333MHz (rated latencies) 2.93GHz/9x333MHz FSB;
  
• PC3-1520MHz (maximum overclock) 3.04GHz/8x380MHz FSB.

Intel Core 2 Extreme X6800 (operating at 2.93GHz - 11x266); 2x 1GB Corsair 9136C5 Dominator with DHX; Asus P5K Deluxe; Asus P5K3 Deluxe; Nvidia Geforce 7900GTX; Seagate 7200.9 200GB 7200RPM SATA 3Gbps hard disk drive; OCZ GameXStream 700W power supply unit; Windows XP Professional Service Pack 2; DirectX 9.0c; Nvidia Forceware 93.71 WHQL, Intel 8.2.0.1014 Inf Drivers;

Memory Bandwidth

Memory Latency

DDR3's better prefetch allows it to keep up with DDR2 at 800MHz, despite the increased latencies. When you scale DDR3 in excess of DDR2 speeds that's when it comes into its own, despite the very lax 9-9-9-24 memory timings. We managed to only get up to 1520MHz (760MHz DDR effective) and unfortunately not the 1600MHz we were aiming for due to current chipset limitations. Future products should provide extra chipset straps that enable the use of DDR3-1600MHz memory, but currently only 1333MHz is available as the top end, and even then you need a 1333MHz Intel Core 2 E6x50 CPU (which aren't available to buy, yet - Ed).

Corsair's DDR3 memory overclocked to 1520MHz was 26 percent faster than 1066MHz DDR2, however and even 1333MHz DDR3 is 14 percent faster in raw bandwidth. But does the synthetic scenario translate into a real world performance advantage? Let's find out...