What's better for AM2- lower RAM latency or faster raw speed?

[996GT2]

I have been trying to OC my OCZ Gold 2 GB kit for a while now, since it is DDR2-667 and most sites recommend DDR2-800 for optimal performance. What it comes down to is this:

My MSI K9N platinum, with the newer 1.2 BIOS, only has 4 settings for RAM speed- 400, 533, 667, 800. At my current overclock of 2.66 GHz, the 533 divider yields just over 600 MHz and the 667 divider yields a RAM speed of over 750 MHz.

At 600 MHz RAM, I can get my RAM to pass many cycles of Memtest86+ at 3-4-4-8-1T timings (2.1V)
At 750 MHz RAM, I can get my RAM to do the same at stock 4-4-4-12-1T timings at the same voltage.

My question is, which setting would yield better performance?

EDIT: I also noticed this-at DDR2-600 (4-4-4-12-1T at 1.8V), my RAM gets HIGHER memory bandwidth than at DDR2-600 (3-4-4-8-1T 2.1V) within Memtest 86+ V1.6. The difference is roughly 150 MB/S. What's up with this?

 

[eplebnista]

curiosity bump

 

[Furen]

DDR2 600 at 3-3-3 should give you better real latency than DDR2 750 at 4-4-4.

 

[TheRyuu]

Technically DDR500 @ 3-3-2-7 (or almost any latency) is faster then DDR400 @ 2-2-2-5 so faster speed is almost always better unless your running some insanely high latencies IN THE REAL WORLD.

 

[the Chase]

Go for the speed.

 

[DrMrLordX]

Since you can run your RAM at both speeds with stability, why don't you conduct some benchmarks on your own machine and post the results here?

 

[A554SS1N]

From the reviews I've seen, it's better to get the higher clocked DDR2.

 

[formulav8]

Just a suggestion, it appears that AM2 likes the longest amount of tRas Cycle Time as it can get (To a degree). So, raise your TRas instead of lowering it and see what you get.

Ex: Instead of 4-4-4-8 or 4-4-4-12, try 4-4-4-14 and 4-4-4-16 and so on until it gets unstable.



Jason

 

[996GT2]

Seriously? Where did you get this information?

I was always under the impression that tighter timings were better...

 

[fbrdphreak]

Originally posted by: 996GT2
Seriously? Where did you get this information?

I was always under the impression that tighter timings were better...

Different architectures respond to different latenies differently. IIRC S939 A64's liked a tRAS of 10 or something, lower timings would reduce performance very slightly.

Personally I wouldn't bother going all crazy, it will make like a <1% difference for all that time you spend tuning.

 

[theprodigalrebel]

I just spent 5 minutes trying to dig up that FiringSquad article on "What memory bandwidth/latency is the best for AM2 performance?" Their search function sucks and googling local site: firingsquad.com didn't help either.

I recollect this: going from DDR2 800 @ 5-5-5-15 to DDR2 800 @ 4-4-4-12 offered a 1-2% improvement on average. Going from DDR2 800 @ 4-4-4-12 to DDR2 800 @ 3-3-3-9 yielded a similar 1-2$ improvement (on average). Some apps show no improvement, some show up to 5%..depends. This was done with the $400+ Corsair RAM.

They declared DDR2 800 @ 4-4-4-12 the sweet spot, though I'm not sure it's worth the usual $50 markup it carries over CAS 5.0 RAM (when comparing in the same manufacturer's product line). I can't link the article now but I'm pretty sure DDR2 1000 @ CAS 5 had better results than DDR2 800 @ CAS 4.