Benchmarks
Corsair's own internal testing seems to suggest to that pure RAM MHz is often more important that latencies. They state that running 3-4-4-8 latencies at DDR 467 (PC3700) gave, on average, a 4.9% increase over benchmarks performed at 2-3-2-6 at DDR434 (PC3500). We sincerely hope that overall CPU speeds were kept constant during testing, so as to not colour results.
As mentioned on the previous page, we'll be putting latencies vs. MHz to the test. Low latencies, in the form of 2-2-2-6 at DDR400, will be compared to DDR467 memory at 3-4-4-8, with CPU speed being held relatively constant. In essence, we'll be comparing a stock 3.0GHz 'C' CPU to a pseudo, overclocked 2.6GHz 'C'.
SiSoft SANDRA first.
SANDRA seems to favour extreme latencies over pure MHz in its unbuffered memory benchmark. That's with a 32.5MHz FSB and memory speed deficit. One may ask why 2-6-2-2 latencies were used, as there's no current Corsair module specified to run at those settings at DDR400 speeds, but many users have found that PC3200 TwinX LL has no problems at 220+ FSB with 2-5-2-2 timings.
On to Pifast next; it's a lover of bandwidth and low latencies.
Extremely close here. Greater theoretical bandwidth is offset by the poor turnaround speed of the high latency RAM. It allows the low latency comparison RAM to just nip in front.
Gaming and 3DMark 2001SE.
A case where bandwidth wins over latency, just. A faster FSB and faster memory speed is better than super-low latencies in this instance.
Unreal Tournament 2003 Demo Flyby is a hair's breadth faster with higher bandwidth, but that's probably more do with the extra 19MHz of clock speed.
Quake III sees it the other way, though, and gives victory, albeit by the smallest possible margins, to the low latency memory. What does all this mean. We'll attempt to explain just what's best for you on the following page.