DFI Admits 5Volt Vdimm Problem

Quote:

---

Originally Posted by jiff

Only problem I see is IF its stated as a DFI NF4 problem, how come I know of three incidents of DFI NF3 users with cold boot/dieing sticks.
My finding...... starts as a cold boot problem then ends up a dead stick.


My :2cents: I think this will only be resolved when the maximers comes out and sticks continue to die, then at least it will rule out the M/B once and for all

---

If the problem comes from the DRAM not able to handle this kind of high voltage for a long time ... There must be a lot of user kill their DRAM using DFI NF4 because it's the only board currently available with easily aceess to high dram voltage :D ... But this does not mean that killing VX DRAM only exist in DFI NF4 board :nono:

Maybe we need to collect more information and ask if there is any other case that user killed their VX DRAM in other plarform + some DDR booster or 3.3V DRAM mod but not with a DFI board ...

By reading what Oscar Wu wrote and Angry Games' conclusions, you will see that what I said might be very true... Winbond-UTT is weak and might not be as good as an old BH-5, throughly tested by Winbond in high-heat/high voltage owens before it leaves the fab. Most memory module manufacturers test sticks manually using special testers but only fabs have owens that heat up ICs to 60-70C and apply high voltage to simulate months of "normal use".
Like Angry Games stated, 99% of truble cases come from users with UTT based sticks, which are NOT tested in a fab and as some may have a long, high voltage life, many others may not.
That was my last :2cents: on the subject ;)

Quote:

---

Originally Posted by mcbarnet007

Oh, why don't you have the bios limit voltage from 3.3v to 4v when jumper set to 5v? that way user will be forced to use 3.3v and not 3.2v

---

Good point and should be implemented.
Besides, the only warning i have seen regards running lower volts with 5v jumper, was posted here on XS by Oscar. How would anyone guess that? I believe that this particular ability should have been mentioned in manual.. It would save many users who missed that post learning about it the hard way.
And lots of good RAM too..

Quote:

---

Originally Posted by Badong

Good point and should be implemented.
Besides, the only warning i have seen regards running lower volts with 5v jumper, was posted here on XS by Oscar. How would anyone guess that? I believe that this particular ability should have been mentioned in manual.. It would save many users who missed that post learning about it the hard way.
And lots of good RAM too..

---

"
ATTENTION:
The DRAM voltage as originally supported by the spec is up to 2.85v only. Jumper J17 allows you to select beyond the defined spec. Although the function is supported (at the hardware level), selecting a higher voltage may cause unstable current supplied to the system board which may subsequently cause damage to the CPU and DIMMs

"

This is what I read from the manual I got , I think it's clear enough ... If you did not need to access voltage higher than 3.2V , why change the JP17 ? :confused:

Quote:

---

Originally Posted by bachus_anonym

By reading what Oscar Wu wrote and Angry Games' conclusions, you will see that what I said might be very true... Winbond-UTT is weak and might not be as good as an old BH-5, throughly tested by Winbond in high-heat/high voltage owens before it leaves the fab. Most memory module manufacturers test sticks manually using special testers but only fabs have owens that heat up ICs to 60-70C and apply high voltage to simulate months of "normal use".
Like Angry Games stated, 99% of truble cases come from users with UTT based sticks, which are NOT tested in a fab and as some may have a long, high voltage life, many others may not.
That was my last :2cents: on the subject ;)

---

Did you killed your VX memory with MSI NEO2 and DDR booster ? What's the voltage you set to kill your VX memory ?

Quote:

---

Originally Posted by wtz54321

This is what I read from the manual I got , I think it's clear enough ... If you did not need to access voltage higher than 3.2V , why change the JP17 ? :confused:

---

Nahh. The point was that running anything below 4v starts putting stress on regulator, the bigger the difference => bigger the stress.
As said already, it's better run 3.6v instead lets say 3.2v. Anyways, it's a simple foolproof and could be easily implemented...

And enough quoting that statement.
I my eyes, it's in manual not because the issue on topic. Just a general warning regards running your RAM above spec. voltage - so and unexperienced user won't attempt frying his Hynix-whatever based RAM with 4 volts..
:slapass:

Quote:

---

Originally Posted by bachus_anonym

By reading what Oscar Wu wrote and Angry Games' conclusions, you will see that what I said might be very true... Winbond-UTT is weak and might not be as good as an old BH-5, throughly tested by Winbond in high-heat/high voltage owens before it leaves the fab. Most memory module manufacturers test sticks manually using special testers but only fabs have owens that heat up ICs to 60-70C and apply high voltage to simulate months of "normal use".
Like Angry Games stated, 99% of truble cases come from users with UTT based sticks, which are NOT tested in a fab and as some may have a long, high voltage life, many others may not.
That was my last :2cents: on the subject ;)

---

I agree with you completely. SOME Winbond UTT can't handle high heat for a long time underload. This is my personal experience I've killed 1 stick because I didn't put a ramsink on it. Right now I have HUGE copper heatsinks on my UTT right now and 90mm fan blowing at them with 3.35v at 263mhz. I've been running this since the beginning and no problem at all.

Quote:

---

bachus_anonym

By reading what Oscar Wu wrote and Angry Games' conclusions, you will see that what I said might be very true... Winbond-UTT is weak and might not be as good as an old BH-5, throughly tested by Winbond in high-heat/high voltage owens before it leaves the fab. Most memory module manufacturers test sticks manually using special testers but only fabs have owens that heat up ICs to 60-70C and apply high voltage to simulate months of "normal use".
Like Angry Games stated, 99% of truble cases come from users with UTT based sticks, which are NOT tested in a fab and as some may have a long, high voltage life, many others may not.
That was my last on the subject

---

Bachus, I think you're probably right about that after all (and you've been saying it for a while too! :toast: ).

You're right, the testing standard for UTT from what we're told is just that, untested... but older BH-5 had all been tested to a standard of operation, and the chips that didn't make the cut were immediately tossed into a garbage can.

Standards being - Part can run X speed, with X timings, at X voltage, with X temperature, with stability for X amount of time.



Electron migration however, is an entirely different and quite scary beast. It was a problem with older Socket A Processors, older intel P4 chips, chips before that era as well. I'm told that the copper interconnects used in modern processors either slows down the process, or lessens the degre of damage. It is also possible that the latest generation of processors have not been overclocked/overvolted long enough for electron-migration deaths to begin occuring.

Electromigration increases the resistance of the metal interconnects within our transistor-based hardware. The transistors inside of our transistor-based hardware have a VCC (VCC = voltage high signal - "VDIMM", "VCORE", "VDD", etc. etc. - stands for a "1" in binary) tolerance rating - once this tolerance is exceeded, the VCC signal can be mistaken for VSS (voltage low signal, 0V - stands for a "0" in binary), or just fubared when the transistor is 'overwhelmed'. When interconnects within our transistor-based hardware have an increased resistance, it can mess up the VCC signal, throwing it outside of transistor tolerance. This usually degrades an overclock slowly over time, with he result that more and more VCC is required to maintain previous frequency overclocks.

Electromigration can also cause broken interconnects. That means dead hardware.

Oskar said that the transistors (and as such, the interconnects) within his "dead" memory were fully functional. It makes sense though, and electromigration could be accellerated in the lower-quality ("unlucky") UTT BH/CH memory ICs as a result of poor (NO) standards testing at the fabrication plant.

Until now , we can only assume that two possible reason that cause the UTT VX dram dead quicky

1 . Unexpected High Voltage or spike in DFI NF4 board to kill the DRAM
2 . Electromigration effect

If it's case 1 , there should be no other board + dram mod or ddr booster kill the dram since they are not powered by 5v rail as DFI NF4 ...

If it's case 2 , there should be some other MB to cause the the dram dead after a time period ...

I think we need to collect more data about what kind of dram being killed , and what's the MB kill it , also the voltage set to kill it( in DFI NF4 or DDR booster or DRAM mod in other board)

Then we can see what's the most possible reason to cause these dram dead ...

We know that it isn't situation #1. A DFI rep just said that the boards do not kill memory, or spike VDIMM.

That's enough for me, to rule out #1.


That leaves electromigration and memory integrity.

Migration on two week old Ram, I don't think so :slapass: In fact, I'm quite surprised to see Oskar mention this as a possible cause.

Put me booster in, switched back to 3,3v line and its working fine. I never had problems (using the 5v rail from day 1) but like this better for a while :D

Quote:

---

Originally Posted by felinusz

We know that it isn't situation #1. A DFI rep just said that the boards do not kill memory, or spike VDIMM.

That's enough for me, to rule out #1.


That leaves electromigration and memory integrity.

---

Well , that's what DFI rep says , maybe there are someboards did have this spike voltage that he did not know ... So we need to collect more data before going to the final conclusion ... Maybe both happened ,,, who knows :confused: :confused:

Hmm, anyone know of CH-5 dieing on this board. I do agree that the ram is dieing due to high volts. I mean any voltage over 3v is nothing to joke with. Think to back in the day when the onyl way to get these volts was a mod. When we killed ram back then did we blame it on the motherboard? We blamed it on the ram or high voltage/poor cooling.

I dunno. After spending much time here over the last 5 days I'm leaning towards the theory that there is an issue with the 5v feed (other than the use of those higher voltages).

Read this if you haven't already.....
http://www.bleedinedge.com/forum/showthread.php?t=10555

I sure hope not because my CH-5 loves volts :(

Quote:

---

Originally Posted by AMD-me

I sure hope not because my CH-5 loves volts :(

---

that's fine, just stay on the 3.3volt feed and either use a DDR booster, or a PSU with adjustable rails to raise vdimm above 3.2volts (using this 'mod' http://www.xtremesystems.org/forums/...ad.php?t=61632).

So i can use up to 3.2v using the stock bios? All i need is 3.2v thanks :D

erm. overclocking, for most of us DFI NF4 owners, has been extremely tempermental with the "Stock BIOS" (1/25), you should seriously consider flashing to either 316, 331, or 510...
http://www.techpowerup.com/articles/other/114 (just rename the BIOS file to NF4LD310 and then you can use the Diamond Flash Utility (I've done it twice now, and works beautifully).

Quote:

---

G H Z

Migration on two week old Ram, I don't think so In fact, I'm quite surprised to see Oskar mention this as a possible cause.

---

Yeah, you're completely right about that. Electromigration, in a processor for example, typically takes a long time (months-years) to manifest itself, and that's with a huge overvolt and high temperatures.

It also doesn't kill hardware immediately. Typically, it manifests itself through a very slow 'degradation' of your overclock, with more and more voltage required to maintain stability, because of the increased interconnect resistances.

It's possible in 6 month old memory... but two week old memory? Preposterous, as you say.

I think Oskar is implying that the electromigration effect is somehow messing with the RAM's ability to be recongized, and boot with an overclock. He mentions that the memory he tested was fully stable in memtest86 - complete transistor functionality. The issue seems to lie with... memory detection, or memory VCC signal strength. *shrug*

Angry_Games' personal opinion appears to be on the UTT memory as well.

Quote:

---

Angry_Games

UTT memory is based on the Winbond CH-5 die, which is a .15micron process. The true BH-5 dies are .175micron...a lot thicker (tho the numbers might not seem like much).

the thinner CH-5 die could possibly be 'weaker' when it comes to stress, voltage, and heat and cause either electron migration or the module to fail completely.

BH-5 with the thicker die doesn't seem to have this problem (which is why 99% of the modules that seem to have failed are UTT, which is an untested CH-5 IC). Also keep in mind that the module might not really be dead...but changed (from heat/stress/voltage) to an extent that the original default properties are no longer valid (ie it might have booted @ 2.6v/200Mhz but now it requires 3.0v @ 200Mhz to even boot).

again, keep in mind that these are my personal theories and are not scientific in any sense (since i am not an engineer of any sort, just a dude who is logical and likes to read a lot and hates being stumped by a problem).

---

http://www.dfi-street.com/forum/showthread.php?t=12687


You've got to give the DFI reps credit. I feel that they have handled this extremely well, in a very professional manner.

ddr booster works fine on the board, ram is in yellow slots, booster in dimm2, bios 510-2 set at 2.5v dimm, 4 pin power strainght to mobo, and using the booster for the regular 20 pins. bios tracks the voltage increase i do with the booster even tho set at 2.5v.

Electromigration does exist BUT......Can someone explain why my Rams 5 x BH-6 and 1 x BH-5 working FINE since year 2002-3 ........NEVER BELOW 3.35Vdimm.....WHY my 6 Rams DIDN'T loose NOT even 1 MHz overclockability ALL these years......ALL of them working most of the time 24/7.......4 of them real 24/7@3.35Vdimm and @ 3.45Vdimm and other 2 for benching AND crunching also (SETI - D2OL) 24/7 @ 3.45Vdimm......Did I got the "golden samples" ?.......I don't think so ......ALL of them are Kingston VALUE Rams and were bought about 30 - 35 Euros each..... ;)

Where is MY electromigration?.....PLEASE give me some electromigration too, so as to see how it is/fills.....Thanks.......! :D

EDIT: my 2 quorters.......He,he,he.......All of you have given some cents in this thread, so I desided to give something too..... :D

Quote:

---

Originally Posted by hipro5

Electromigration does exist BUT......Can someone explain why my Rams 5 x BH-6 and 1 x BH-5 working FINE since year 2002-3 ........NEVER BELOW 3.35VDimm.....WHY my 6 Rams DIDN'T loose NOT even 1 MHz overclockability ALL these years......ALL of them working most of the time 24/7.......4 of them real 24/7@3.35Vdimm and @ 3.45Vdimm and other 2 for benching AND crunching also (SETI - D2OL) 24/7 @ 3.45Vdimm......Did I got the "golden samples" ?.......I don't think so ......ALL of them are Kingston VALUE Rams and were bough abour 30 - 35 Euros each..... ;)

Where is MY electromigration?.....PLEASE give me some electromigration too, so as to see how it is/fills.....Thanks.......! :D

---

You're using BH-5/BH-6, not totally untested Winbond UTT.
I have 3 sets of 2x256MB BH-5:
First (Mushkin LvLII) have been used in my NF2 for over 16 months now with 3.5v constantly, as NF7-S is vio=vdimm modded.
Second (Kingston) was in my Neo2, abused with a Booster and 3.7v for 6 months.
Third, (Mushkin LvLII) in DFI NF4 and earlier in Neo2. 6hrs a day, 3.8vdimm for the last 5 months...

No problems whatsoever so far... I think it's because BH-5 was throughly tested in fabs before module manufacturers bought it. Winbond UTT has not been tested at all by fabs...

I had 2 sets of UTT... Both went belly up after 2-4 weeks of using it on Neo2.

It must be those greek gods looking down on you my freind....;)

Topboy