Micron D9 and high voltage, electromigration effects

[mad mikee]

Learned my new important thing for teh Day

Guys

Its not all about voltage...DRIVE is also real important, to much will do more damge than voltage ever will.

If you have a board with drive tweaks keep the drive as weak as you can keeping stability....the more drive you apply the more current you push thru the dimms and the more damage you do.

[STEvil]

Can electromigration be stopped or slowed down by cooling? (water or subzero)

Water, not really. Sub-zero might get you somewhere.

Drive strength is quite literally the amperage provided @ the specified voltage. You can look at it this way...@ 400mhz, that's 400cycles/second.

Let's say you get punched 400 times, by an smurf. Annoying, but it isn't gonna hurt you.

Now let's replace that smurf with GodZilla. Let him hit ya a couple hundred times. Nothing left of ya but mush, eh?

lmao, great way to phrase it!

[ziddey]

My experience is the opposite Mark. 10gb of dead D9 and counting, alot of it never saw above 2.2v. All active cooled with 120mm, 3 different boards.

Good theory/explanation btw Micutzu.

lol ouch????


edit: what does saaya mean

[Raja@ASUS]

you can always check for heat on the dimms at low voltage to gauge how strong the manufac in question keeps the drive, hotter @ low vdimm means strong drive. Worse is putting 2.5vdimm thru a set of ddr2 then runnning countless hours of orthos to test stability, that sucker deteriorates ram fast when you have high drive and high vdimm simultaneously. It is nice to be given vdimm drive strength options, also look out for variance in drive between bios releases, there are sometimes changes..

[ineedaname]

I have a feeling that a lot of heat spreaders on ram contribute to killing them.

A lot of the heat spreaders use really crappy thermal pads that do more harm than help.

Even good heat spreaders seem to only help for certain ranges of temperatures. After a point the thermal pads just don't transfer enough heat.

I also agree with Gautam, I'd bet that bad voltage regulation kills more ram than electron migration cuz there are plenty of ppl killing ram with very low voltage which doesn't make sense at all.

Seems like 680I boards kill a lot of ram. I bet that their reference vdimm regulator is really crappy, IE the evga board. Which makes me wonder what temps my vdimm voltage regulator on my striker is.

I still remember when 7900GT's first came out and a good bunch of them that used the reference design all died because of crappy voltage regulators that were just too hot. Just cuz an IC spec says it can run at 90C doesn't mean you should let it run at 90C. This goes to show that nvidia's reference design uses the cheapest parts possible.

[STEvil]

ineedaname - Bad voltage regulation causes fluctuations which will result in increased electromigration, they are complementary.

As to voltage regulation, if you want it good go pick up a server board. My X7DAL-E has lots of hardware for controlling voltage for up to 24GB of ram

[coop]

Does someone know if changing the voltage to either mems or cpu by means of a VR, change the ratio of current? For example, setting vdimm to 2.1 in bios and adjusting VR for real volts of say 2.6. Is that different than just setting to 2.6 in the bios? Also conversely, lowering the volts from a higher setting in bios.

[cadaveca]

depends on mem VRM design. BUt yes, potentially.

[ineedaname]

ineedaname - Bad voltage regulation causes fluctuations which will result in increased electromigration, they are complementary.

Yes I realize this but that's not what I was pointing out. I mostly meant bad voltage regulation as in it jumping to 3V and just frying the ram totally in one go. I guess you thought I meant it showing up as 2.3v when its really a constant 2.5 or so.

[STEvil]

no, I meant varying like a yo-yo.

[Revv23]

commando must regulate vdimm better than p5b deluxe. i killed 6 sets of D9 on my P5B, with the same benches i've killed none on the commando.

in my experience P5B had terrible Vreg on the ram, when i switched to hipro maximizer i needed much less voltage for the same speeds...
(like from 2.4v to 2.15 - yes that big, the abit quad GT was much better..but i still think the maxi is best.)

[fakhrain]

and by an interesting coincidence, lowering drive strength at high voltages on my dfi helps stability... by the time i've over 2.5-2.6v, it is around level 3 to level 7, weak

On my DFI nF-M2, the DRAM output driving has value of 1~4. Does lower number means weaker drive strength?

[Roger_D25]

Thanks guys for all of the information in this thread, very informative read to say the least! I actually thought I knew what "drive strength" was back when I was running DFI boards but after reading this I realize I really didn't! I wish it was easier to measure current, like it was for measuring voltage!

Although I realize I can't adjust the drive strenth on my motherboard I would still be interested is learning how to figure out how various motherboards compare as far as drive strength goes (for motherboards that don't have drives strength adjustment that is)? I assume this is not possible thought, right?

[superray]

This thread is really helping.

[ozzimark]

On my DFI nF-M2, the DRAM output driving has value of 1~4. Does lower number means weaker drive strength?

yeah, lower is weaker.


edit: unless it's like thr ultra-d and the nf3-250gb, then i *think* 1 and 3 will be weaker than 2 and 4, weak vs normal drive, with the lower number being lower.

[superray]

yeah, lower is weaker.


edit: unless it's like thr ultra-d and the nf3-250gb, then i *think* 1 and 3 will be weaker than 2 and 4, weak vs normal drive, with the lower number being lower.

What's different about NF4 ultra-D?

[eva2000]

in my experience P5B had terrible Vreg on the ram, when i switched to hipro maximizer i needed much less voltage for the same speeds...
(like from 2.4v to 2.15 - yes that big, the abit quad GT was much better..but i still think the maxi is best.)

weird i had no difference in max ram clocks with P5B Deluxe 0804 and Maximizer up to 2.45v for my DDR2 modules, only >2.45v got me higher ram clocks limited by 1.65v vMCH being available.

Did this on my maximizer last night heh

[funkflix]

Nice thread and great informations here!

No problems here so far, running my GMH up to 2,6xV for short benshing sessions and 2,2 - 2,3V 24/7. Vdimm looks really stable to me, no drops or something, cooled by XTC Memory Cooler.

[ahmad]

To add to the first post, designing ICs in a way that reduces electromigration exists of course, but that leads to a few things:

1- Greater latencies
2- Insusceptibility to greater voltages
3- Price of manufacturing

Which sounds a lot like some other non-micron chips. I think crucial is taking the easy way out and giving us modules that work at higher voltages, but not engineered appropriately in the beginning to prevent electromigration. Of course heat and voltage/current fluctuations play a huge role, but nothing that can't handle a bit of heat and some minor power fluxes should pass QC anyways.

So I guess that leaves me with the conclusion that d9s will be great for benching, but hardly for impressive 24/7 clocks that will last more than a few months.

[justin_c]

Well, see I ran my rams at 2.3v. From a P5B deluxe. And it died in 1 month. Do you think the "jitter-free" current from a ddrmaximizer would prolong its lifetime?

[Raja@ASUS]

'jitter-free' - 'lower ripple', a lower ripple would in theory prolong the life of on board capacitors and to a smallish extent the ic's;- on a volt for volt basis. Obviously we don't have hard facts either way for running low ripple supplies on ddr2; just theory.

[afireinside]

Ah so higher drive strength means more current... Explains why my HZ gets hotter than hell at 2.28v on my xbx2...

P5B is a complete POS. I toasted a set of ballistix @ stock on them.

[Raja@ASUS]

afaik, at the performance point we now have available, electromigration will exist in all products running at these speeds when overvolted past spec, due to necessary die shrinks to make the speed available in the first instance. Smaller die sizes mean narrower, more fragile traces that need to be packed tightly to conform within universal sizing specifications, and also give the speed capability (engineering wise) to remain competitive. How long ram will last on various boards when overvolted is still in the air as drive strengths are not always known...


regards
Raja

[Micutzu]

Guys

Its not all about voltage...DRIVE is also real important, to much will do more damge than voltage ever will.

If you have a board with drive tweaks keep the drive as weak as you can keeping stability....the more drive you apply the more current you push thru the dimms and the more damage you do.

True, drive strenght can do damage. But, AFAIK drive strenght refers to the voltage amplitude on the data channels (DQ0 to DQ15) and may be tweaked to ensure stability and signal integrity in any conditions. As the transistors inside the RAM cells have only ON/OFF modes, doesn't really matter what voltage you are applying to the gates to open them, or at least not directly (only the leaked current will increase, wich is a small part of the total drain-source current).

[Revv23]

weird i had no difference in max ram clocks with P5B Deluxe 0804 and Maximizer up to 2.45v for my DDR2 modules, only >2.45v got me higher ram clocks limited by 1.65v vMCH being available.

Did this on my maximizer last night heh

Well like i said on my quad GT the maximizer didnt make nearly as much difference. it could just be that i had a bad P5B and you have a good one...

nice clocks BTW