Freaky said:
Just curious. How do the items in RED affect memory performance ? And which is the best in benchmarking memory - Sandra or Everest ? Coz' the difference in values in each is substantial enough for me to ask this. Also, the difference in bandwith in both S/W between XP MCE and XP Pro x64 in my system is substantial. Which should I take into consideration ?
http://img342.imageshack.us/img342/7372/a64tweak3mz.jpg
I've provided some very good links for tweaking RAM here:
http://www.techenclave.com/forums/showthread.php?t=4849&page=1
In particular look at the DFI nf4 BIOS Guide. That should help answer your questions.
I'd suggest though that you shift to A64Tweaker version 0.6beta. I like that better. Generally, with ValueRAM you may not have much leeway in deviating very much from the SPD. Nonetheless trying to tweak the settings may still be a worthwhile effort
. Note that the net effect is a sum total of all the parameters, and by losing some in one, you could potentially gain with some other. This will be most conspicuously evident as you're fine tuning when the RAM is at its limits
.
A very rough sketch of those parametric effects below:
Trc, Trfc:
Supposedly, Trc = Tras + Trp.
Both Lower for P(erformance) and higher for S(tability). Roughly, for Trc 7-higher and for Trfc 12-higher.
Tref:
This is a b**chy parameter to tweak. Supposedly the value is dependent upon HTT and, consequently, RAM speed. But in my experience the behavior is not monotonic. Generally, 15.6us are aggressive and 1.95us are for stability. But very high refresh rates may be counter-productive. I'd suggest starting with 3.9us and try to go either way and each time run a 1M SuperPi simulation to check the effects. In my experience, at very high RAM speeds I tend to go toward 1.95us for Stability, and toward 15.6us for Performance; mostly I've had good success with 3.9us.
Max Async Latency and Read Preamble:
Lower for (P) and higher for (S). For 2 x 512 sticks most certainly, at the very minimum, 7/8ns for Latency and 5-5.5 for preamble for stability.
Idle Cycle Limit
Lower for (P) and higher for (S).
For BH-* sticks 16-64 and for TCCD 64-* seem to be good for performance and stability.
Dyn. Counter
Enable for (P) and disable for (S). Works in conjunction with Idle Cycle Limit value above.
Between Everest and Sandra, stick to one while tweaking. Sandra is good for Winnies - bcoz it gives better bandwidth over Venice
! If you're interested in examining the effect of latencies on bandwidth you could use the unbuffered option in Sandra. How to do this is IIRC provided in Eva's OC-guide in the link above.
HTH
