FSB:DRAM

[EarthwormJim]

Originally posted by: n7

Originally posted by: EarthwormJim
N7, try playing with the tRD phase adjustments now. You may be able to run a tRD of 4 on some phases, if not all.

In that menu, setting each option to enabled does a -1 to your current tRD. Do each one individually so you can find out which ones you can and can't enable.

If you don't see that option in memory timings, try a newer bios my original bios didn't have it.

On my DFI board (same as yours) I can't post at a tRD of 6. However with the phase adjustments, I can set my tRD to 7 and with all phases enabled, I'm really running a tRD of 6.

I have lots of questions for you regarding this mobo...
It's been actually really frustrating to work with, since there's almost too many options, & maintaining stability seems to be insanely hard.

Truth is, i haven't been able to find a single combination of settings that can pass HCI Memtest.
I know it's because of having 8 GB, but i would have thought that if i cranked NB voltage i could get it to at least make it to 100% thru four HCI instances, but nope.
Yet i can pass P95 custom set to stress all 8 GB easily.

It seems HCI puts too much stress on the NB...board just can't hack it.

I don't know why you wouldn't be able to post at TRD 6...likely you need more vNB...
I mean, if i can get it running with four 2 GB dimms raping the board, i woulda thought anyone can.

FWIW, i'm using the 266/800 ratio if you hadn't already done the math to figure that out.
The 333/667 & 333/800 ones seem to hate me unless i pour in vNB...

Anyway, maybe we need to make another thread here for discussion on this board.

Honestly you may be pushing your DDR2 800 just too far. I had some Micron D9gmh and it wouldn't really do anything above DDR 1000. My g.skill 2x2gb ddr2-1000 won't do anything above ddr-1100 without HCI failing. You can try bumping up your memory voltage, but I wouldn't go over 2.2v actual (as in what your measuring software says) for longterm.

And yes this board can be extremely frustrating while you're trying to learn what each setting does. Check out the official thread on it on xtremesystems. Also the anandtech review can be helpful.

http://www.xtremesystems.org/f...4898&highlight=dfi+p35

Prime blend doesn't really test ram out that well, I always used HCI. It's pretty fast at finding errors too.

As far as not being able to post at a tRD of 6, I think p35 boards use an equation to check your tRD setting before they will post. Not necessarily because of instability, but just cause.

Using the equation from the anandtech tRD article, a tRD of 6 isn't valid. 450fsb ( a bit higher than what you're at), 333:800 strap and a cas of 5 at ddr 1080.

 

[myocardia]

Originally posted by: EarthwormJim
You can try bumping up your memory voltage, but I wouldn't go over 2.2v actual (as in what your measuring software says) for longterm.

His RAM has Powerchips IC's, not Micron. 2.1v is the maximum they should ever be given, at least on a continual basis.

 

[lopri]

350 FSB with 266/800 ratio?! Isn't that 2:3? (meaning.. your RAM is at DDR2-1200)

Originally posted by: myocardia
His RAM has Powerchips IC's, not Micron. 2.1v is the maximum they should ever be given, at least on a continual basis.

Yes. And I think these ICs are picky on chipsets. On a P5E-VM HDMI (think of it as a P35 variant), it does DDR2-1066/CL5 with just 1.82V.. and that's the max. lol. It won't go any higher no matter what I do but It has no problem running at that speed with less than 1.9V (counting some overvolting). It might show different characteristics on different chipsets.

 

[lopri]

My thinking with regard to ratios on Intel chipsets is, at least for now, that we should go with per-chipset basis, or even per-board basis. Different vendors are using different technique in the BIOS, and they are not always uniform in their behaviors with clocking. Add straps (or manual tRD if a board allows) to that and it becomes even bigger mess. While there is a clear tendency that higher bandwidth will yield higher performance other things being equal, the other things are rarely equal from my experiences. What I noticed with P35 is that while the overhead from running ratios (other than 1:1) is much less severe than previous chipsets, but at the same time things became so forgiving (i.e. loose) and the performance gain from memory clocking is even less pronounced than before. Said that, if you have a board that gives a lot of control and actually follow through with your control (I noticed that ASUS boards rather ignore my commands and go with the safe route while tweaking), you can probably find the sweetest spot with these boards. But it will certainly take some time to find that spot.

I am still advocating 1:1 (or 1:2 for DDR3) due to the least overhead (waste). If you think about 975X or P965 - you will remember that these really liked only one or two ratios other than 1:1. Particularly 4:5 for 975X and 2:3 for P965. 4:5 on P965, although the absolute frequency of memory would be lower than that of 2:3, just didn't work well. Throw in 4 sticks and things get a whole lot worse. Everything seems to point to the inefficiency of dividers on Intel chipsets, even with the latest chipsets that are much more forgiving when it comes to ratios. Using dividers has an inherent disadvantage, IMO. There is a gain, yes, but the gain < cost.

AMD CPUs tend to not have this problem thanks to its IMC (although it's common to see a later stepping with better memory controller than its previous one) and performance responds more linearly to memory performance.

 

[n7]

Yes, obviously things will vary depending on chipset & hardware combinations used, but in general, more bandwidth rarely hurts things on Intel unless timings have to be severely loosened to achieve it.

Doesn't necessarily help much either, so it will depend on what you're running for software, what goals you have, etc.

Not sure if it was the same for all P965 boards, but when i tested 1:1 vs. 4:5 on my P5B-D, i found 4:5 better, no question.
Unfortunately, Fusetalk somehow killed the old thread, & i don't really care to run more benches on my P5B-D + Celeron E1200...heh, or maybe i do, just not right now

I also found it was incredibly hard to run 4:5 with 4 dimms...actually basically impossible when higher FSB was used, since the board didn't allow nearly enough subtimings alteration i think.
I couldn't even post with a decent config 4:5 with four dimms on there, but could do 2:3 okay.

On this P35 i have now, running the 1066 strap's ratios vs. the 1333 strap's ratios seems to be easier on this board, & due to being lucky enough to have a 10x multi on this chip, i don't need to worry about high FSBs, which allows for me to go for 2:3 again.

It might be completely just me being insane, but i've always found 1:1 to be a horribly boring ratio to run.
Too easy & default & normal...
Yeah, likely just me insane.

So i've always preferred trying for those nice high clocks & bandwidth vs. timings @ 1:1, especially you can never get anything lower than 4-4-4 with four 2 GBs aside from extremely low speeds.

 

[lopri]

It's hard to have fun with 4 sticks.. In my case I'm just glad that it works.

And you're not alone in thinking the boredom that is the name of Intel overclocking. Basically you find the limit of your CPU and you're left with one or two usable dividers.. It'd be a different story if there were some more multipliers to toy with, but the only way to have that is, well, paying up through your nose. Ugh.

P.S. Do you know what these are?

http://forums.anandtech.com/me...=2163872&enterthread=y