Benchmarks
Remember that three different combinations of FSB and memory speeds are being benchmarked with roughly the same CPU speed. The basic premise is one of determining if pure memory MHz can help offset the performance losses incurred by higher latency RAM.
SiSoft SANDRA's unbuffered benchmark first.
As is the case at Hexus, the benchmark was repeated a number of times to validate findings. A look at the XMS3700's results shows that the 3GB/s+ score obtained by the XMS4000 modules isn't internally consistent. We can simply put this down to the fact that the XMS4000 TwinX memory kit was benchmarked with the supplied 1GB set, whilst the comparison two were both 512MB sets.
How does Pifast see it ?.
That's a little closer to what one would expect. Extra bandwidth, albeit it with a slower CPU speed than the TwinX 3700 set, gives the best performance. We'd imagine that a refrigerant-cooled 3.0C could hit 250FSB+; perfect partners for these high-speed modules.
Pure bandwidth is 3DMark's friend. An overclocked 2.4C, with the XMS3700 TwinX in tow, would comfortably outpace a standard 3.0C with low latency timings, according to 3DMark 2001SE.
Quake III seems to agree, too.
If we'd increased the TwinX XMS4000's CPU speed to 3040MHz, as is the case with the 3700 set, the former would have an even greater lead. How much of that lead is attributable to like-for-like bandwidth and not to the 1GB test size ?. Difficult to know, really. Bottom line is that dual-channel PC4000 memory is almost perfect if your setup can bring out the best in it. That necessitates running an Intel-based machine with, preferably, a highly overclockable 800FSB CPU and either a Springdale or Canterwood motherboard.
