FSB vs. Memory speed

[XeNO6696]

I've read some conflicting info on the 'net, mainly that you want the memory speed to be a multiple of the FSB speed, my FSB is 1333 so the way I read it is DDR2=667 or 1333 would be the best, but I've got mine running at 1100 and at least by Sandra, everything appears to be benchmarking faster.

So what gives? What's the *real* scoop on this? should I loosen up my timings and target an FSB speed of 1333?

 

[MyLeftNut]

This is how it works, 1333 is 4x multiplied. Meaning that your ram only runs at 333mhz (DDR2 667). So if you're running 1:1 ratio 333mhz quad pumped will be 1333mhz FSB.

For simplicity, just run it at 1:1 with the tightest timings which remains stable.

 

[XeNO6696]

Only problem here is that I bought 1066 memory, not 667, and running at 1:1.66 is the only way I can get close to 1066 (1100 actually). System runs stable, there's no problems, but I was just wondering (specifically) if there's a penalty I'm paying somewhere else by running the system asynchronously. Or is faster memory always just that--faster memory?

My bandwidth as it sits is a little over 6000MB/s, but if I dropped the system as a whole down to 1:1 and underclocked the processor so that it all matched at 1066 would this let me see an overall increase in speed? Or is where I'm at pretty much as fast as it'll go?

Note here that gaming is only a secondary concern, this system is primarily a DAW and so I need ability to move and crunch large amounts of data more than tight timings.

 

[MyLeftNut]

How high does your cpu overclock to? With this in mind, what multiplier and memory speed are you running?

For example, I have:

CPU: 8x
Memory: 450 (Memory has SPD for 1066, but I'm only at 900)
Mem/FSB Ratio: 1:1 (450)

Total: 3600mhz


I don't know how much memory latency past 4-4-4-12 will effect the kinda work you do, but I find that usually, having the latency at 4's is the best compromise between speed and latency, that way you can have both. As for the work you do, I would think the best way is to benchmark it yourself with different settings either running asynchronous with FSB ratio higher than memory with looser timings, synchronous with tighter timings, or synchronous with lower multiplier and higher memory speed with looser timings (if required).