Intel Latency Dilemma

After a long period of burn in, my KVR 333Mhz 512MB memory module is now able to run at 400Mhz with full stability in conjunction with my KVR 400Mhz module, which was underclocked for this slower module. This, however, leads me to a new dilemma, stay at slower speed for better latencies, or increase speed?

The current numbers I'm working with:

333Mhz @ 2.5-3-3-7
400Mhz @ 3-4-4-9

This is at standard voltage. Now, I know this is very subjective, but "which is better?" I'm a gamer primarily, but I also do some DVD encoding and other memory intensive activities. Is there a benchmark available that can show what's really happening?

Note: This is the first rig in my sig.

The last Intel CPU I owned was a 50MHz 486DX2, but I have the impression that with Intel, you still need to run the RAM synchronously with the CPU. If I've got that right, and increasing the RAM frequency also means a higher CPU frequency, you'll definitely get more performance from the 400MHz option.

yea, but intels have memory dividers too so im pretty sure you can run like 4/5 or somthing with the memory faster, or 5/4 and vice versa. 200/166 and ect. on this, you really need to do the benches to find out yourself. latencies DO help, but you probably arent going to notice them too much unless your really looking for them in benchmarks. memory bandwidth is effected but shere mhz increase that more favorably.
do a test.
3dmark01 your machine at the two frequencies and latencies to see which one gives you better results. same core speed of course, just change the memory divider.
I would think 2.5-3-3 would be superior. you never know though, you might be able to stuff some volts into that kvr to get it running those timings @200 mhz too. i have a sammy that can do that up to 220 mhz @2.9volts.
cheapest ram around too heh.

I had the impression that using memory dividers with Intel CPUs caused a fairly serious performance hit, but I could easily be wrong.

On my intel system (3.0ghz prescott) changing from 3.75ghz to 4.0ghz and changing the memmory to a 5:4 resulted in a 150 point lower 3dmark01 score for me. 1:1 ratio with ram at DDR500 speed on my system, beat the 5:4 ratio 270FBS (ram at 432mhz). Even with tighter timings. Maybe I am unlucky?

It all depends on what your bandwith is. Running asynch has its merits but it depends on your RAM having low latency