Interesting research on dimm's

[UnreaL]

I have been doing alot of testing of my Twinmos UTT on the DFI Ultra-D and I have come to an interesting conclusion.

I have 2*512mb sticks and I have tested each stick individually on dimm#2 (orange futherest from cpu) at various fsb, volts and timings.

As an example I decided to use these sets of results:
With 1 stick I could run 200fsb 2-2-2-5(-9-15-2-2-1-3) with 3.0v and get 0 errors while looping memtest #5 for 8 hours. This applied to both sticks when used individually..
When I use both sticks at once running 200fsb 2-2-2-5(-9-15-2-2-1-3) with 3.0v using dimm#2 + 4 (orange) I got several hundred within 12 loops.
This at first made me think that my PSU/motherboard was to blame, however I decided to test varying the voltages and so increased the voltage to 3.1v and it has now just looped 8 hours at 200fsb 2-2-2-5(-9-15-2-2-1-3) error free.

From this it is fairly each to come to the conclusion that the board/psu is not supplying enough volts when using 2 sticks but I wanted to confirm this with others so if it is possible maybe for others to try and do a similar set to see if there is a point were individual sticks are fine but paired they are not.
I am fairly sure the PSU is alright as it is a Tagan 480w so based upon the same hardware as the much aclaimed OCZ series. This means it is likely to be the DFI causing the problem, it is using the 2/18 bios and I will update to 3/10 tomorrow but I was just interested in if this was a common problem that not alot of people have spotted yet.

Does anyone else have similar results?
Lemme know! :toast:

[Gogeta]

What cpu were you testing the dimms with? There is a possibility the on-die mem controller could be causing the errors as well. I'm not saying this is the definite cause, just an additional factor to consider. If its an FX cpu, I'd say the possibility is next to zero, but it could be a different story if its a winchester.

[HiJon89]

Its well known that running dual channel decreases the overclockability of RAM, it puts more stress on the memory controller.

[UnreaL]

Im using a 3200+ winchester so maybe thats it... but then again why would volts fix it if its the mem controller?

[conrad.maranan]

If you give the CPU more voltage, it's natural that you're also giving the on-die memory controller some voltage as well.

[MaxxxRacer]

he uped the mem voltage, not the cpu voltage conrad

[Millyons]

It shouldn't be voltage distribution but for some reason on all boards some slots are better than others.........in the nForce2 days i was wondering the same thing and measured the voltage each slot is getting in the slot itself and they were all the same

[conrad.maranan]

My bad. It's been a long day. :yawn:

[UnreaL]

hmm thought there was something confusing about your post :P

[UnreaL]

Hmm it appears that this stuff is highly temperature sensitive, I ran all my tests with an XP-120 and 60cfm fan blowing onto memory + 70mm fan actively cooling mem. Also the top mosfet was cooled by another 70mm fan.

It appears that by putting a side on the case I get more errors and if I leave it without a sidepanel I get less!

I have 2x 80mm + 1x 92mm air in + 2x 80mm air out + Tagan psu fans so I am suprised it makes such a difference!

[persivore]

I don't know if this is true for new RAM interfaces, but on a very old board I was using (486 using 30 pin SIMM) I found that some of the lines between the RAM modules were connected together (data lines iirc). If this is true for DDR RAM/A64 then adding more RAM to a system will increase the capacitance on each of the pins that is connected together, which could decrease the overclock ability. Increasing the voltage will help to improve signal strength and help to compensate for the extra capacitance, so the overclock ability should improve.

There could also be more ripple/voltage fluctuation being introduced to the Vdimm supply, so decreasing the overclock with 2 sticks of RAM.

[UnreaL]

Anyways it does 200 2-2-2-5 @ 3.1v but im going to give it that 48 hour burn in at 3.2v and see what happens

[JuanFlaiter]

I don't know if this is true for new RAM interfaces, but on a very old board I was using (486 using 30 pin SIMM) I found that some of the lines between the RAM modules were connected together (data lines iirc). If this is true for DDR RAM/A64 then adding more RAM to a system will increase the capacitance on each of the pins that is connected together, which could decrease the overclock ability. Increasing the voltage will help to improve signal strength and help to compensate for the extra capacitance, so the overclock ability should improve.

There could also be more ripple/voltage fluctuation being introduced to the Vdimm supply, so decreasing the overclock with 2 sticks of RAM.

I like that theory. :)

[Millyons]

Here is a pic from an old thread here, these are measuring points for each dimm slot, for who ever wants to reaserch more on this topic.........my system is in a case, i don't have a second system atm to play with.

Oh ya its easier to measure with ram sticks that don't have heatspreaders on them

[Nubius]

Its well known that running dual channel decreases the overclockability of RAM, it puts more stress on the memory controller. Does that mean it'd be worth while to put it in a non DC setup and obtain a higher overclock?

[Gogeta]

Does that mean it'd be worth while to put it in a non DC setup and obtain a higher overclock?

Well, if you had a 3700+, you wouldn't be dissappointed :D. One dimm is also much less stressful on the mem controller, similar to comparing 2x256 to 2x512 in a dc setup.

Only problem is running 2x512 on most boards.

[JuanFlaiter]

Here is a pic from an old thread here, these are measuring points for each dimm slot, for who ever wants to reaserch more on this topic.........my system is in a case, i don't have a second system atm to play with.

Oh ya its easier to measure with ram sticks that don't have heatspreaders on them

OT: Whats ATM?

[conrad.maranan]

ATM = at the moment

[Millyons]

like conrad said...............

one other thing.... even with single sticks of ramm dimm slots have different oc ability, even with same voltage, so this "issue" is probably just cause of the mobo traces

[enzoR]

its jsut causeing more strain on the northbridge and the signals may be interfereing with each other. thats why increaseing vdimm helps

[persivore]

I checked if the data lines on the dimm slots are conencted together, and some of them are.