Benchmarks I
I'll start off with SiSoft SANDRA's memory benchmark. Although this isn't particularly useful in differentiating between memory timings, it does give us some reasonable information on bandwidth efficiency in an ideal world.
DDR-300 is simply memory running synchronously to the FSB of the processor, 150MHz in this case. Both the Corsair and Samsung modules are set at 150MHz, 2/6/3/3 timings. I expect little difference from this result, and that is what I get.
DDR-400 is achieved by the 3:4 CPU:DRAM ratio. The Samsung PC2700 module is run with CL2.5/6/3/3 timings but shows no adverse results because of it. SANDRA is a lover of pure MHz memory speed, almost irrespective of timings. 3GB/s, on this benchmark, is perilously close to the RAMBUS scores laid down at 133FSB. DDR is certainly coming of age.
How does this theoretical bandwidth translate into real-world performance ?. I'll run Pifast and find out. Pifast is a simple program that calculates the constant Pi to any number of decimal places you choose. I've chosen 10 million places. Pifast has previously shown a distinct liking for fast system memory, bandwidth is key.
Have a look at those results with the knowledge that nothing apart from memory speed has been changed, the processor has remained constant at 2550MHz (150FSB). Does memory speed matter ?, it certainly does in this benchmark. We're looking at < 5% gains from just raising our memory speed from DDR-300 to DDR-400. The gain from going from CL2.5 to CL2 manifests itself in a little over 0.2s lead, hardly substantial.
Next we'll turn our attention to MP3 encoding. We're benchmarking by encoding a 638MB custom WAV file into 192kb/s MP3 using the LAME 3.91 encoder and Razor-Lame 1.15 front-end. Does memory bandwidth play a pivotal role here ?
It appears not, a one second lead is nothing to be concerned about, it seems as if this is strictly a CPU affair.
