Interpolating memory timings at different frequencies?

bberson

Distinguished
Oct 25, 2006
363
0
18,780
Is there a sensible mathematical formula or consistent method that can be used to interpolate memory timing when altering frequencies, or is it all just seat of the pants guesswork?

Say, ignoring EPP profiles or in their absence, if money burned a hole in my pocket and I have PC2-6400 rated for 5-5-5-12 at 800MHz, and I choose to use it at 533MHz, is there a formula that tells me what timings I could use or should use?

Something tells me it isn't necessarily linear.

Also, I'm a little perplexed as to how the MEMORY FAQ describes "400MHz @ CL=6, tRCD=6, tRP=6" as translating to "5-5-5".

Finally, if a stick is advertised or described merely as "CL=5" are there any assumptions we can make about tRAS?

Thanks for any clues!
-Brad
 

rwaritsdario

Distinguished
Aug 2, 2006
3,017
0
20,780
Is there a sensible mathematical formula or consistent method that can be used to interpolate memory timing when altering frequencies, or is it all just seat of the pants guesswork?
I cant speak for all the modules in general because some can have very different potential than others. Some are also factory OCed to the max, while others have too much potential to be measured.

Say, ignoring EPP profiles or in their absence, if money burned a hole in my pocket and I have PC2-6400 rated for 5-5-5-12 at 800MHz, and I choose to use it at 533MHz, is there a formula that tells me what timings I could use or should use?
Always aim for the lowest possible. Again, I cant generalize.

Finally, if a stick is advertised or described merely as "CL=5" are there any assumptions we can make about tRAS?
If they only advertised the Cas then the other timings arent as good. They could have force the Cas down by keeping high on all the other timings.
 

Mondoman

Splendid
The underlying timings are typically fixed times. However, since timing specs are reported as clock cycles (the length of which changes with the clock frequency), you can pretty much linearly interpolate values. The main thing to watch out for is the granularity of the reported values (i.e. a 5 clock timing just means "more than 4", as you don't know whether it's 4.1, 4.5, or 5.0).