nF4 timings: No changes in speed? How come?

[Radical_53]

When clocking my nF4, I've played around with all those memory timings and settings. Some friends of mine, who had the board earlier than I did, gave me some good suggestions, settings they had tested out.

I tried some of the settings posted here.

 

Thing is: The only option that's really fast for me is to load optimized defaults on bios 310p and change only CPC to enabled. Any other changes result in less performance... I simply don't get it.

 

Anyone with a clue? I roughly get 7550MB/s in Sandra 2005 at 270MHz, I think it's not as fast as supposed to be (still the fastest setting, as I said).

 

Suggestions are greatly appreciated.

[Poisonsnak]

Well some timings like CAS latency really should have a big effect on your bandwidth. But many of the timings (if you have ever played with alphas on the nF2 you will know what I mean) do in fact reduce your mem bandwidth in order to allow you to increase the clock speed.

 

Then again it might just be how your RAM behaves, if anyone else has used OCZ ram maybe they can comment? I'm a (bh-5 wannabe) ch-6 guy myself.

[Radical_53]

Yes I know, the better known latencies do have great impact on performance. Thing is, that my OCZ are rated at 2.5-3-3-8 anyhow, bank interleave is enabled with the optimized timing and I always set the CPC/1T command by hand to be sure.

 

I was just curious as many others were talking and messing around with all the other timings, I just never saw what the effect was, you know? Only to tweak the settings for a slightly higher frequency, remaining at the speed they were at before?

[Smoken Joe]

I only had very small improvements after changing the first 4 settigs to optimal. I also keep bank interleave enabled too. 2.5 3 3 8 for me also.

[Radical_53]

Interesting. Good thing to know. Saves me a lot of time

 

So, you say *small*... could you call some numbers on it? So then, it really seems as it was only good to gain the last few MHz when your CPU can do more than your current multiplier would allow

[Smoken Joe]

Small as in you could only see a few percentages of change in the artafical benchmarks and a diffrence so small you cold not even see a FPS diffrence in real games.

[Radical_53]

A few percent? Well, that would actually be quite something If it was something like 2-3%, I'd take a look at it.

To be honest, I couldn't even find such a difference by raising the HTT like 25-30MHz. 250x11 is ~ 1.5% slower than 275x10. So some percent would be quite much Or comparing TCCD to BH-5... 2.5-3-3-7 compared to 2-2-2-5. The difference, at the same frequency, is roughly 2.5%. As TCCD can achieve higher clock speed, the difference is even smaller in real life, maxing out both types of sticks.

[Smoken Joe]

The artifical bences like sandra where it really means squat in any real use. Unless you consider benchmarks and specificly sandra benchmarks or the like an inprotant part of your life and not a tool dont bother too much use them to fix problems after you have a nice stable memory oveclock. I was also not thinking improvement either- settings that were not so good(but stable enough to boot) vs the best.

They are not very far off at all stock.

[Radical_53]

One thing I'd consider would be 3D 2k1 (for a benchmark), if those timings showed an improvement there, I would have thought about it. But no other benchmark, there's only one thing I need the speed for: Games. No FX, no SLI and no need to be the allover benchmarking king

[Radical_53]

Wel ok, I couldn't leave my fingers off those timings... still the same result.

 

I tried bios 310, 310p and 330-1, 310p remains fastest.

 

Then, I tried some sets of timings. The standard "optimized" setting, which is this (beginning upper left, A64 tweaker 0.6):

 

2.5-7-17-3-2-8-3-2-1-2-200(15.6)-1-16x-128bit-disabled-disabled-7-7-5-256-disabled-200-enable-normal-disable.

 

Strangely, this setting is faster then the tweaked setting:

 

2.5-7-12-3-2-7-3-2-1-2-200(3.9)-1-16-128-dis-dis-7-7-5-16-ena-200-ena-normal-dis.

 

 

 

I simply don't get it. It's not that the tweaked setting wouldn't be "much faster", not even by a small margin, no: it's slower!