Cachemem is a benchmark that is influenced quite a bit by memory latencies. This benchmark really tells the tale of what I believe DDR2 can provide for us. At 533MHz memory speeds and 200MHz fsb speeds, the scores are quite poor in my opinion. As the bus scales though and we reach DDR2 667MHz, the higher speed of the memory is able to offset the rather lax timings of the Crucial Ballistix PC2-5300 DDR2.
Sciencemark doesn't seem to be quite as effected by memory timings, but does continue to illustrate our point described above. As the DDR2 memory speed scales, the scores seem to get much better overall.
Aida32 seems to highlight this memories potential better than any other benches run. Scores here are definitely good demonstrating the ability of the Crucial Ballistix to provide some big-time bandwidth.
Sandra scores at default CPU and motherboard speeds are quite poor. At 200FSB the memory is running below its default speed though at only 533MHz, as opposed to rated speed of 667MHz. Once we are able to scale the FSB and consequently the Memory speed up to 240+ we see scores jump quite a bit. At 250MHz FSB the memory is at its rated speed and bandwidth is quite good for a Pentium 4 platform, but one does wonder where the potential 8,000MB/sec motherboard(at 250MHz fsb) and 10,880MB/sec memory bandwidth is escaping too. Granted, memory bandwidth cannot run past system bandwidth, but there is still quite a void between theory and results we are seeing.
To calculate potential bandwidth-- 128bit bus(2x64bit dual channel)/8(bit to byte conversion) x frequency(after DDR applied) = bandwidth in MB/sec
Onto our thoughts and conclusion ->