Question about performance when underclocking ram

[scarywoody]

Specs in sig.

I just installed all my crap and have been kinda itching to OC a bit. I'd like to keep my FSB:RAM 1:1. To do that I have to unerclock my ram to 266 mhz.

Just for a small taste I put the FSB at 300mhz and OC'd to 3.0GHZ.

Hooray my first OC.

My question is....which will perform better?

2.66ghz with ram at 400mhz

or

3.00ghz with ram at 300mhz

I guess I'm trying to find the line where ocing the fsb will surpass the performance of having 400mhz ram. Make sense?

 

[scarywoody]

Oh and temps stock were high 20's idle and low 50's Loaded.

3GHZ were low 30's idle and mid 50's loaded.

Using TAT

 

[almerac]

im willing to bet the 3GHZ will be faster, few things are really limited by ram bandwidth

 

[scarywoody]

Crap now that I'm thining about it I wonder if I should reduce the multiplier to get my ram up to 400mhz if I can't with the multiplier at 10. Hmm shouldn't someone have done all this by now

 

[NotAPimecone]

Crap now that I'm thining about it I wonder if I should reduce the multiplier to get my ram up to 400mhz if I can't with the multiplier at 10. Hmm shouldn't someone have done all this by now

People do it all the time.

I've heard that lowering your multiplier on a P965 board overclocks the northbridge core clock. This would go a long way towards explaining why the 965 board did so poorly in the recent 680i vs 975 vs 965 article at Tom's (they could only get the 965 board up to 350MHz on an X6800 with the multiplier dropped from 11 to 7) even though lots of people (myself included) run much higher FSB's on 965 boards.

However, 10 / 8 * 400 = 500MHz, and the board should be able to handle it.

http://www.thetechrepository.com/showthread.php?t=30

 

[scarywoody]

Try both and run a few Benchmarks to see witch is better. I'd be interested to find that out myself. As my motherboard sucks i can't find out for myself. The motherboard just won't allow a stable OC. Thats a Gigabyte GA-K8N SLI-Pro for you.

What's a good benchmark that does a good cpu and ram test. I've only really played around with 3dmark.

 

[RichardJSampson]

im willing to bet the 3GHZ will be faster, few things are really limited by ram bandwidth

I'm going to jump in on this thread, because I've been wondering about this myself. I'm 100% stable at 3.45 (345 FSB), but with 400 MHz 4-4-4-12 RAM, I too, would like to be running my FSB at 400 Mhz or higher.

So I was intrigued that almerac indicated that bandwidth speed makes a neglibligle change in overall performance.

Can anyone corroborate this? I have this fancy new DDR cooler (the Thermaltake Cyclo DDR cooler) I've yet to install it, so if getting bandwidth of 400+ with a lower multiplier gets me *substantially* more performance at the same CPU clock speed, I'm all for it.

NotAPimecone said "people do this all the time." With what results? Having a girlfriend and a 70hr+ job makes it hard for me to screw around with benchmarks often. So for those of you who have, and presumably there are at least a few of you, what have been your results?

 

[scarywoody]

I messed around a little last night.

PCMARK05
Score stock (2.66GHZ, RAM 400mhz)
Score was 69XX
3dmark06 was 69XX

Score (3.33GHZ, RAM 333MHZ)
Score was 78XX
3dmark06 was 62XX

Only voltage change was the cpu to 1.4 per OC guide sticky. Ram timings were 5-5-5-10 as listed on the ocz gold memory @2.0V. I'm not really sure when or why or if I should change the timings ???

Temps were high 50's, so I called it a day. I put everything back to stock since based on the 3dmark scores. I might try OCing this weekend without running a 1:1 ratio. I need to transfer all my crap into the antec900 case. Should improve my temps quite a bit from the crappy old school case I have now. I'll probablyt start over and redo everything since I didn't get the exact pc / 3d mark scores.

 

[NotAPimecone]

You only have to change the timings of you're running your RAM faster than its rated speed.

You only ran your RAM (rated for 5-5-5-10 at 800MHz) at 800MHz and 667MHz, so no timing adjustments would be necessary.

Now, if you raised your FSB above 400MHz (and therefore your RAM speed above 800MHz), or otherwise raised your RAM speed with different ratios, then you would probably have to relax your RAM timings at some point.

I get away with 850MHz at 4-4-4-12 on my RAM which is rated for 800MHz at 4-4-4-12, but if I go much higher, I have to relax my timings. I don't know how high over rated speed your RAM will run without relaxed timings, of course. Test extensively

 

[scarywoody]

OK I messed around and have some results. I just started using pcmark because it was taking to long to run that and 3dmark.

vcore set at 1.4 and mem set at 2.1V

Multiplier FSB:RAM CPU(Ghz) PCMARK05
10 2:3 2.66 6938
10 1:1 3.33 7734
10 4:5 3.33 7955
9 4:5 3.15 7644
8 1.1 3.2 7731
10 4:5 3.5 8229

So I ran into blocks trying to go higher with the 8/9x multiplier.

I think I might persue the 4:5 with 10x multiplier.

I tried to get to 3.66ghz to get a qghz Oc for fun, but I couldn't get it.
I cranked up the mem voltage to 2.25 and set the timing to 5-5-5-15. Should I try cranking up the vcore a bit? Not sure all the steps to go through once I hit a block.

Anyhoo temps at 3.5 were 42ish idle and 68ish full load.