Benchmarks
We'll start off by looking at what kind of unbuffered bandwidth each combination has, according to SiSoft SANDRA's benchmark.
The unbuffered benchmark does take latencies into account to some degree. Notwithstanding the high latency of the OCZ PC-4000 DC set, its pure DDR500 speed puts it on top of the pile, as expected. It even overcomes a slight CPU MHz deficit when compared to the low latency set.
Our resident Pifast test, conducted to 10m places and now copied on many a website's, is a decent indicator of the MHz vs. latency trade-off. The OCZ comes out on top, but not by as much as one might expect if viewing the SiSoft SANDRA graph directly above. The higher latencies cause the OCZ to take a little longer than expected to push out the data demanded by the CPU. Wouldn't some DDR500+ 2-2-2-5 memory be lovely. We live in pure hope.
3DMark 2001SE v330, though, is partial to pure DDR MHz than low latencies. A monster 1GHz quad-pumped FSB and synchronous memory takes over 500 marks out of the low latency combination, which, let's remember, is running that little bit faster on the CPU.
Things tighten up under Quake III's bandwidth analysis. Standard DDR400 memory, run at 3-4-4-8 timings, doesn't stand a chance.
Once again, A higher FSB and high latencies are able to overwhelm a low latency DDR400 combination. Running synchronously at high FSBs, even if that high memory MHz is accompanied by weak RAM timings, is a sure-fire method of hitting the benchmark heights. It's no wonder that a certain DigitalJesus, equipped with some mind-numbing DDR550-capable OCZ PC-4000 memory, is able to run amok on 3DMark 2001SE's leaderboard.
