Originally Posted by Manyak
I believe it's only SB-E where anything past 1600MHz is useless, due to it having a quad channel memory controller to make up for it.
It depends on workload and architecture....
Originally Posted by 8800GT
Hmm definitely an interesting article. His calculation of latency in NS is certainly interesting. By his calculation my Mushkin blackline set to 1600mhz @ c8(10ns flat) should theoretically be better than 2133mhz @ c12(11.25ns). but my Ram at it's standard 1600mhz @ c9 also gets 11.25ns. So maybe this proves that instead of going for faster speeds and then lowering the timings the reverse is better? my brain is so scrambled right now after reading that haha. Anyway, i ran some memory benchmarks and found that although my overall latency is lower, 2133@c12 was still theoretically faster than 1600@c9 AND c8.
You can't just look at those numbers. You have to consider workload and architecture as well.
Lower timings are better for random while higher frequency help with sequential.