ok..i KNEW i am not alone :)Quote:
Originally Posted by EMC2
Btw. so it looks like the OCZ booster may fix this ?????
Printable View
ok..i KNEW i am not alone :)Quote:
Originally Posted by EMC2
Btw. so it looks like the OCZ booster may fix this ?????
Hey my findings are circumstantial and nowhere as scientic or accurate as that scope, only true test is for someone to scope the booster's supplied vdimm/vtt fire up occt and take a look
I’m new to all of this, don’t take my word .. I’ve just fought that prime95 stable from day one, and would love to have a solid, yes that’s it, or no, back to tweaking answer to my personal journey/mystery
Chad
yes i am waiting for someone replying who has scope readings using the booster.
Also...i know that i know FAR less than those guys in that thread here....but i had a few thoughts....and i saw some postings [EMC noted he saw drops while the video initialized] which made my thoughts even more interesting:
<> monitoring VTT *NOT ONLY* when under load with OCCT/Prime95.
WHEN does this happen ?
I say this because ever since (and still) i get occasional BSODs eg. while i am literally doing "nothing"...just with FireFox open or browsing thumbnails on HD.
There *might* be situations where the vtt drops down other than OCCT/extreme load. Just an idea.
<> they already said there's a relationship "drive strength" <---> vtt drop "extent". I wonder if there is a relationship "VTT drop" <--> "Vdimm voltage".
I am wondering this because right now i am experiencing this extremely bizarre fact that my TCCD needs 3.1V and is not stable anymore at eg. 2.8 or 2.9 i had it before. (same bios settings of course).
Is the increased VDIMM "somehow" compensating or influencing the drop of VTT ? (I say this because i cant explain otherwise why my TCCD want 3.1V, i never heard of TCCD wanting more than 2,9 or so...but it's the case here)
Just my $0.2
my tccd needed atleast 3v to do anything...week 449. but that doesnt explain why you were stable at 2.8 and need 3.1 now...Quote:
Originally Posted by flexy
fwiw, based on emc2 comments in various places, I did back my drive/data strengh down ... I got my dfi-street oc database stable entry (and 24 hr prime stable here) with 6/2 - I had been running this at 8/4 (8/3 in later bios)
Chad
tony at ocz did a simple vdimm mod
http://www.bleedinedge.com/forum/sho...80&postcount=1
and he claims this mod gives very good results, eg. where he needed 2.9v/3.0v before (TCCD) he can now do with 2.6 !!!
He doesnt have an explanation, but one of his theories is it COULD provide better VTT with that mod. <--- ? (Which needs to be proven w/ scope which i dont have !)
ALso..from what i READ now the ddr booster does NOT solve VTT issues ! Still.....it would be interesting to see how the ddr booster does (scope/occt).....if it does NOT provide better VTT then it must be doing something ELSE, which, in term leads to a cleaner VTT.
(And here we are again with my speculation whether there is some relationship VTT <--> VDIMM voltage).
This is one of the MOST interesting threads btw.
good find but it is the same thing as Malves mod in this thread (1st post) im pretty sure. basically connecting pin 6 to ground, malves used pin 4 for ground while Bigtoe used the ground on JP17 block.Quote:
Originally Posted by flexy
see--> http://img8.echo.cx/my.php?image=themissinglink7fv.png
if pin 4 is not ground then its not the same but id think it is
I don't think that a DDR booster will have any effect on the stability of the Vtt supply. AFAIK the booster only supplies the Vdimm, and doesn't effect the Vtt supply, unless the Vtt regulator on the motherboard tries to follow the Vdimm set by the DDR booster.
Maybe this explains why hynix or other generic value rams which usually requires stronger drive strength doesnt work well on this board?
Very interesting thread indeed.
Well, the Booster needs to affect Vtt, since it is altering Vdimm and Vtt must TRACK 1/2 Vdimm. Vdd Vddq Vref and Vtt all must track correctly. On DDR main ram Vdd and Vddq are just connected....as opposed to GDDR2/3 on video cards were they are seperately controlled to precisely nail maximum ram speed.
So Vdimm(Vdd/Vddq) and Vtt(Vtt/Vref) are DIRECTLY related at all times, unless you have a sucky DDR voltage regulator design, like the NF7-S did, where Vtt did not correctly track, and required a seperate Vdimm and Vtt mod to achive high Vdimm and correctly tracked high Vtt.
On the TCCD issue, you may be right, that the Vtt is the root of why higher Vdimm is needed. You end up raising Vdimm to the point where the "MAX LOW VALUE" on Vtt is high enough for the ram to be happy. That fixing this droop would allow a much LOWER Vdimm for TCCD is likely.
The scope doesnt lie, that droop we are seeing is real and it needs to be eliminated. Perhaps the calculations are wrong and the Vtt load is alot higher...on a transient/instantaneous basis...and the design needs some help.
I asked about the booster because it would have to generate a bump on Vtt too, and maybe setting the booster to match the Vdimm/Vtt of the motherboard would effectively DOUBLE the output current rating of the regualtor circuit NET, since both MB and Booster regulators would be feeding the Vdimm/Vtt bus's. Alot like getting another 9173A and soldering it on top of the original doubling the output.
improbving the Vtt regulators output with caps as well would entail adding a 560pf, a .1uf, and a 2200uf electrolytic to cover all spectrum of signal transients.
Remember, yes a 560pf is fast enough to respond to the transient edge, BUT it contains so little charge it cant actually make up for the deficit in instantaneous current needed at the 0's to 1's transistion in Test #8, for example.
I believe the 9173A's will run OK in parallel, so I guess we need to experiment with soldering in a second one. Some AS Ceramic goop on the bottom 9137 then piggy back, bend the 2nd one's pins down and solder it inplace over the original. Find a spot farthest from the regulator output...like one of the .1uf bypass caps for the Vtt resistor arrays, and solder the 2200uf electrolytic there, add the 560pf/.1uf ceramic caps to the regulator output pin. (each one cap to Vtt, othersie of cap to a good ground VERY cloe to the cap).
Yes, it is the same vdimm mod. Pin 4 that I use is ground. Really doesn't matter what ground you use.Quote:
Originally Posted by stealth17
Hey EMC2,
Maybe it's time to test the board on the scope with the vdimm mod.
well just out of curiosity (and since my name is George :) ) i took my DMM and watched VTT - of course it totally confirms what EMC said. Clearly visible on the dimm.
Start OCCT and you can literally see it crashing on the DMM when the VTT drops down like 10%-15%. Voltage down --> crash :)
If i had the parts i would do the voltage mod just to see what happens then, but i dont have a VR right now to solder.....waiting for results by others.
I STRONGLY tend to say that this is a (previously) overseen board/design flaw because we are so MANY people with inconsistencies in their overclocks and the same issues....so i am very much looking forward for a fix to keep that VTT steady.
Makes a lot of sense !!Quote:
Originally Posted by uwackme
In my case i dont have a doubt that my sys got more and more instable, and as said various times my TCCDs need more and more ram.
Thank to this thread i MIGHT make a conclusion (even as a amateur in those things:) ) that the source of those (my) problems are not the DIMMS itself but has *indeed* to do with that VTT issue. (Inc. VDIMM to lift up VTT <--)
() bad batch of Regulator, board design ? POSSIBLE
But in my case "something" (Regulator ?) seems to get worse....since more and more vdimm is needed.
Q: Is it likely that an voltage regulator "degrades" ? For some reason i see it that if it would be a bad batch then it would stay at a certain "bad" level (ie: instability)....but PROBABLY would not degrade over months (???)
However, the only part i can think of "degrading" over time would be caps, like leakage, which would affect that issue over time.
But then i am only speculating. ALso, AFAIK EMC already concluded its NOT a heat issue etc.
Vdimm mod shocked me quite a bit, ram that needed 2.8V+ at 300fsb now runs well at 2.65v same settings. I will look into VTT and see if the mod helps it along with the booster if i can find the damn leads for the thing ;)
hipro's booster supplies VTT also, so there may be a way of just adding a cleaner VTT to the output leg and see if that gains us some stability...just a theory mind but it does help on the max3 which featured a similar mod i helped develope a while back.
One other alternative would be to tweak the output high for VTT, most dimms will tollerate an overvoltage on VTT its just an under voltage that causes issues, this way we could see a mean VTT value where it should be, high off load and slightly low on load...again just a theory.
I'm really glad to hear that, bro.:) Maybe it does something to VTT?Quote:
Originally Posted by bigtoe
So quick and dirty test before i go to bed, VTT is actually to high on my board and is tracking to a vdimm that is 0.02V higher than it actually is. The vdimm mod had no effect on the droop under load as i saw it with it on and off.
Now my board may be a cherry, i have 3 others here and i will try and hook them up to see if the VTT acts the same on all of them. What we need is an external means of dialing in the VTT, There must be some way of manipulating pin4 to do this.
Hi Tony... yes there are ways ;) And FYI... adjusting Vtt by POT helps a little... but you can't go much because a) Vtt start off high without load already and b) the droops are too large under some conditions. George's new Maximizer would solve it, but it's a tad high $ for just the needed Vtt solution. If you have multiple MBs... would you be willing to rob one for a RT9173A and parallel a pair to verify it's a pure load condition?
But... why I dropped by tonight was to answer the question as to why it might be a good thing to use the Vmem mod to adjust your memory voltage rather than programming it in the BIOS... at least for Vmem above 3.2V.
Here is what happens during a re-boot of the MB... note it doesn't matter if it's a re-boot initiated within windows, pressing ESC during memtest, or hitting the reset switch... in all cases just after the VGA detect and POST tests, the Vmem regulator takes a drop like this:
http://img207.imageshack.us/img207/6...notated4wy.png
This happens with the newer BIOS's that have the "low Vmem during Memory-POST cold boot" fix in them. I don't know about anybody else, but one wouldn't think that would be the best thing in the world. This only occurs if Vmem in the BIOS is set to a voltage greater than 3.2V
Here is a complete picture of what happens during a cold boot now. You can see when the Vmem level is initially set to the value programmed into the BIOS by the user, followed by the same several second dip down to a lower level. You can also see a glitch occuring when that initial level is set:
http://img143.imageshack.us/img143/4...otannotate.png
One last note... Vmem (at least on my MB) is 100mV over the value set in the BIOS for the higher voltages ( more than 3.2V using 5V rail).
Have other info collected, but time is up... work tomorrow.
Peace :toast:
A few things we need to be aware of:
All bios files except the one I released apply 0.1V more than what you set in bios if you set TRCD to 2..this was a fix for some corsair dimms that had a few issues. In my bios i had oskar remove it.So if you are looking to test true vdimm you need to install this Bios and remeasure.
Next to get back to my VTT thinking.If we supplied 9173A direct off the 3.3V line and adjusted the pull down circuit so that we could lower the output quite finely this could cure all our issues. At present Vdimm in Vin and voldrop seems to be the issue causing VTT to drop also.TCCD dimms are real power hungery so we may just be seeing to thin traces etc on the board to cope with the load.
So while i know adding a pot onto pin4 would allow a tweak up it would not allow a tweak down, so we need to mod the atcual circuit dictating the 1/2 divide on the output to allow us to work from 1.25V all the way up to 3.2v.
This is the problem of bios that reset 3.3V above dram voltage control pin back to default before program 3.3V above voltage and cause the dram voltage lower down below 3.3V ... If the system is not hang up during this time period , it's ok ... 7/04 bios can shorten this time period to be within 10ms ... During this voltage transition period , the bios code will not running until the voltage transition is completed and prevent the system from hang up ...Quote:
Originally Posted by EMC2
Connect Control voltage tab to 3.3V will help a little bit ... But after ddr voltage exceed 3.3V , it will perform worse than connect control voltage tab to DDR voltage ...Quote:
Originally Posted by EMC2
Control Voltage=DDR voltage
http://oskarwu.myweb.hinet.net/VDDRCNTROL.GIF
Control Voltage=3.3V voltage
http://oskarwu.myweb.hinet.net/3V3CONTROL.GIF
Original RT9173A performance ...
http://oskarwu.myweb.hinet.net/RT9173A.GIF
Yes, a problem with higher Vmem if you used the 3.3V rail... you could use either the Vmem regulator's first stage output (you said it was limitted to ~4.2V) or the 5V rail.Quote:
Originally Posted by OSKAR_WU
Spec allows for up to 6V on Vctrl:
http://xtremesystems.org/forums/atta...id=33770&stc=1
If you go back and look at the scope shots in post #47, the "regular" drops are 60mV from the nominal center condition, which is 2x the spec'd 30mV as shown in the transient response ratings for the RT9173A.
But the bigger Q is why the 200mV+ drops like shown in post #62 under some operating conditions? Looks like either current limitting is kicking in or the RT9173A is going unstable...
The first dip that occurs during a cold boot is ~10mS long, seen in the 2nd pic in post #117 at the Trigger point. That's where you program the higher Vmem from the minimum level before starting memory detect/post. Here's a zoom in of that area:Quote:
Originally Posted by OSKAR_WU
http://xtremesystems.org/forums/atta...id=33773&stc=1
But why the second drop down to 2.7V? Vmem was already properly set to the level set in the BIOS back at the first point and the complete POST has been run at that level. Why not leave it set as it already was? Particularly when the second time there is an overshoot of almost 200mV...and when noise is added on top of that you get this (note BW for Channel 4 is opened up to 250MHz in this pic):
http://xtremesystems.org/forums/atta...id=33774&stc=1
Thanks for the BIOS Tony, but I measure true Vmem ;) And yes, noticed the 0.1V discrepancy...Quote:
Originally Posted by bigtoe
If you go back and look at the scope pics of Vtt in post #62, you'll see that Vmem doesn't droop at all even under the worst case (200mV) Vtt droop conditions. You can look at the 2nd pic in post #95, which has Vref in it and see that it doesn't droop either (this also shows Vmem isn't drooping since Vref is a simple resistor divider between the pairs of DIMM slots).
Further data point... I looked at Vtt using UTT (CH5) memory last night... Vtt still drops the exact same way and by the same amount. You can see a sample in this thread about 3.3V input levels needed, in the first pic where Vmem was set to 3.0V with the 3.32V input rail ;)
Regarding the POT... you can tweak up and down... just have to use all 3 legs ;) But again a POT doesn't solve the issue of the large scale changes I've seen. IF it is a Vtt-based load condition causing the droops, it is going to require either an amp-based control circuit to feed the Vref input (pin 4) of the RT9173A so you have dynamic feedback to hold Vtt under control or the addition of a) a second RT9173A in parallel to help handle the load or b) a replacement Vtt circuit that can supply the required current (a couple of possibilities come to mind).
Peace :toast:
1 . Use 5V is not recommend by richtek in recent batch of regulator ... Using 4V is one thing that can be consider ...Quote:
Originally Posted by EMC2
2 . The nominal center of actual board is different from dummy load testing like the richtek datasheet ... The patter in memtest8# already cover the full current sinking/sourcing mode of this regulator , I will only check the peak-to-peak range of the result of actual board ...
3 . 94mv in my original measurement is not very good , but acceptable when you consider that the VTT is 1.45V but not 1.25V in the richtek datasheet ... And this is the TCCD/300MHz/1T case , if the DRAM is BH5/CH5 type or the DRAM frequency is much lower , the value will be smaller ...
4 . I don't know what happened to your board , but I have tried 4 different board(different batch/model) ... All of them act very close to the pic I post ...
I don't know if you test this with 7/04 bios ... With 7/04 bios , the period should be within 10ms in the 1st stage and 2nd stage ... I can not explain the detail of all the bios programming voltage transition limitation , it's a protection behavior I set ...Quote:
Originally Posted by EMC2
i hope people *here* dont confuse the VTT droop with the VDIMM droop when VDIMM is close to 3.3 rail ? :) I know they confuse those two issues elsewhere.
VTT droop also occurs with LOWER VDIMM (eg. 2,8 or 2,9) where VDIMM is NOT fluctuating at all.
The "VDIMM fluctuates" is a different story than "VTT droops".
VTT drop: Occurs ALWAYS (?) no matter what VDIMM is at and what rail you grab your vdimm off.
VDIMM drop: Occurs if your VDIMM is too close to your 3,3 rail...eg. starts here once i set VDIMM to 3.0, gets really bad at 3.1 (like EMC showed yesterday)
I have modded an A00 with vdimm and VTT;) i will see what effect this has if any on overclocking etc.
You could use a diode to connect the 5V rail to the Vcontrol input, as this will lead to a .7v DROP in the voltage at the pin...4.3V.
Keep the leads as short as possible. Dont use long leads on things, they act as antenna's to add noise into the circuit.
1. I noticed the note for the ACL5 part, but you aren't using that version are you? And yes... 4V should be fine since Vmem can reach 4V and is thus already used...Quote:
Originally Posted by OSKAR_WU
RT9173 Discussion picture:
http://img221.imageshack.us/img221/4...eakdown9ij.png
2. Yes, the nominal center of the actual board may be different, but it still exists... and the Vtt still goes in one direction due to sourcing current and the other direction due to sinking current... nothing to due with the MB will change that. Looking only at the peak-to-peak variation is misleading... and doesn't take into account assymetrical loading issues. The nominal center by design of the part is Vref+Vos. You can either use Vref as nominal or Vtt @ minimal/zero load. It's pretty obvious when looking at what Vtt does under testing that it does have a real and measurable center position. One example would be it's value during Memtest T5, where it stays rock solid and equal to Vref... another would be with Memtest paused, where you see an apparent small Vos in the positive direction.
3. No it isn't very good... especially when assymetry is taken into account. Regarding the 1.45V versus 1.25V - as shown in the discussion pic I put together from RichTek's spec, the variation of Vtt is almost purely a function of load current, not Vtt level, as can be seen and is annotated in the pic. That's also why the % delta for DDR1 is less than for DDR2. Note that the 3A delta is right at 2X the 1.5A delta as well. From my testing to date, this is confirmed, as I see no appreciable difference due to the level of Vtt/Vmem.
Regarding whether it is related to TCCD or 300Mhz or frequency in general, here's some captures using CH5 based DIMMs at two different frequencies (and lower than the TCCD captures). Note that these were both taken with everything EXCEPT the memory speed kept constant.
240Mhz UUT Memtest T8:
http://img301.imageshack.us/img301/3...notated2tg.png
120Mhz UTT Memtest T8:
http://img139.imageshack.us/img139/8...notated0wn.png
You'll notice that the Vtt magnitude change remains the same between 240Mhz and 120Mhz MemClock... and right at the same as what was seen with the TCCD at 300Mhz.
4. I don't think it's just my MB... but since I've used 2 different PSs, 3 different CPUs, and 3 different sets of memory of 2 different kinds, it is restricted to the MB in general.
I do think you aren't seeing the very large changes that I've recorded because of test condition and equipment differences most likely. Two questions come to mind after further testing tonight...
a) have you used the test conditions in Windows that I sent you the info on?
b) have you run those test conditions with a Winnie rather than an FX?
c) what Vcore levels have you done testing with, just default or higher?
I did go back this evening and also re-checked the "large step" conditions I had previously seen in windows, this timing using 2x512MB of CH5 based DIMMs. The first run I did not see the magnitude of change I had previously. So... I sat down and analyzed all the differences between the two sets of runs.
a) CH5 versus TCCD
b) 200Mhz FSB vs 295Mhz FSB
c) 200Mhz MemClock vs 295Mhz MemClock
d) Vmem 3.3V versus 2.9V
Everything else was exactly the same (ok... temp might be +/-2C different)... including external test equipment, probe points, etc.
So I decided I would eliminate the differences I could easily, one at a time. First thing I did was go into the BIOS, set the FSB back to 295Mhz, dialed in a memory ratio of 4/5 to keep the UTT speed down, and re-ran the test.
Here is 295x9:
http://img139.imageshack.us/img139/8...notated0vv.png
Bingo - we have the problem showing up again. Now, was it the higher frequency on the memory or on the CPU. So I made a series of 3 more runs, each one with everything EXCEPT the CPU multiplier the same:
295x8 Win MemTest:
http://img139.imageshack.us/img139/4...notated0fn.png
295x7 Win MemTest:
http://img139.imageshack.us/img139/8...notated9ly.png
295x6 Win MemTest:
http://img139.imageshack.us/img139/4...notated6ba.png
The last run (295x6) is very close to the first unrecorded run (scope trigger set too low, looking for a bigger droop) where setup was 200x9, 1:1 on memory. Note how the magnitude of the change goes down each time.
This actually makes some degree of sense... one big thing in common with the majority of conditions under which the large scale (>100mV) droops occur are when the CPU and memory utilization rate changes by a large amount and the speeds are relatively high.
The only other condition to date that I've seen cause the large magnitude droop on Vtt is the video switchover that occurs during Win2K bootup... but it is another potential large scale current demand change on the MB and full-bore utilization to low utilization...
It appears now one of two things are going on with the large scale droops - either the regulator is loosing it (stability) do to power demand changes or there is some condition where we get a higher than expected load demand on Vtt due to some unforeseen combination and the regulator's overcurrent circuit is tripping.
I did that testing with 510fixed BIOS. I haven't looked at it yet with 7/04, but I will. Hopefully the shorter duration/method prevents the overshoot I saw (if not may need to raise it halfway, pause, then the rest of the way).Quote:
Originally Posted by OSKAR_WU
hmmmmmm... this "protection behavior" occurs at the same relative time as when the Video is re-initialized (right after VGA detect/POST) and when video modes are changing (my test bed monitor makes a distinct, audible 'click' when video modes are changed)... one of the conditions that I've seen the large scale Vtt fluctuations is during the Windows initialization of the video and video mode change... related perhaps?Quote:
I can not explain the detail of all the bios programming voltage transition limitation , it's a protection behavior I set ...
Peace :toast:
Curious which Vtt mod you decided to try out... and look forward to your efforts :)Quote:
Originally Posted by bigtoe
--- uwackme ---
Just FYI... Vctrl is a :banana: to change... it's on the TAB of the part, so the part has to be totally removed and isolated (plus HS), then put back down to connect up the other 4 pins to make a Vctrl mod :( Someone want to give me a mod-mule MB to use tho...
EMC2 and OSKAR,
I noticed the same voltage behavior with OCZ PC3200 Gold BH-5 (OCZ4001024ELDCGE-K) memory a while back. I no longer see such issue exists with the OCZ Plat. Rev2 TCCD memory. I even posted the question at DFI-Street forum about a week ago. No one responded of course.
P.S. My video card is acting strangely (BSOD every 2-3 hours) in recent days. Any possibility the NF4 Ultra-D board could do something to the video card?
On two occasions after changing bios upto the 704 then restarting after reconfigure (I have done this on hundreds of boards/chipsets/CPU's so no stranger :) ) I have noticed SEVERE screen corruption that looks like the graphics card (X800) is being either overvolted or the pci-e bus is wayyy over speedQuote:
Originally Posted by suio
I have only ever come across such behaviour in the past when I have overvolted due to modding a vid card done back in my 8500 days :confused:
nice job by you guys measuring and all that :)
Btw. do we have that vtt issue no matter what DIMM slots we use and if it's related to load created by the DIMMs, and do we get smaller drops by using eg. only one dimm ?
Just wondering.
Well I certainly haven't studied enough electronics yet to analyse and comment on the graphs and technical explanatios above, but a question to Oskar Wu and/or EMC2:
My CPU died the second I pressed save settings and exit in the 704-2 BIOS. I had changed VID to 1.200, VID special to *133%. This is a setting I had used in previous BIOS's and found gave me the optimal VCORE for my Venice 3000+ 0517 EPMW. The BIOS would read 1.56V, which was perfect for allowing 3ghz prime/bench/stressCPU stable on air cooling.
Now, of course you would understand that I am very frustrated with losing such an amazing chip, and naturally that I would point my finger at the BIOS, so it would be great if knowledgeable people like you could look into the issue and see if the BIOS did introduce some voltage control problem.
I have tested the RAM and graphic card, both are working fine in my friend's system, and my CPU is confirmed to be dead. No power even runs through the chip, since no heat is being generated at all.
I haven't yet tested the DFI NF4 Ultra-D board with another CPU yet, despite having thoroughly resetted the CMOS. I am very apprehensive about sticking another CPU into this board, especially with the 704-2 BIOS still in it. My response from DFI technical support was to RMA the board, and that there may indeed be something wrong with the BIOS... but that's all I've heard, and I've sent an in-depth email regarding the symptoms, to see if they can take the information and use it constructively to debug or verify the 704-2 BIOS, and let me know if the BIOS was at fault or if it was the board, or if (unlikely) it was just something else or some freak incident. I haven't had a response for days, but that's OK, I'm a patient person. I have ordered an Abit AN8 though... don't want to potentially risk other hardware.
This was the original response from DFI after I had briefy explained my problem:
_____________________________
Dear Customer
Thanks for your letter of July 06. According to your problem description, there might be something wrong with BIOS. Please contact with your retailer and apply for RMA service. I apologize for the inconvenience may cause you.
Sincerely Yours truly,
Tech Support
DFI Inc.
Issue NBR: (~~~~snip~~~~)
___________________________
So this is the reply I made, and seeing as there are many knowledgeable people here - maybe you can take it in and see something that makes sense... I'm no electrical engineer yet, so my theories and stuff can be radically impossible, so please take that into consideration. I do however, have alot of experience with hardware, and have been overclocking for a very long time.
___________________
Hello DFI,
Thankyou for the reply. Unfortunately after extensive testing and debugging of various components in my system today I have found that my Venice CPU has died, most likely as a result of a bug in this 704-2 BIOS (Oskar Wu, Hellfire, Merlin version).
I believe the VID control may have a bug, allowing the board to spike the voltage when using certain VID & VID special settings. I believe this is the reason, because usually the commonly experienced symptom of constant over-voltage (typically around >1.75v for 0.09u A64 CPUs) is that the CPU is gradually damaged due to electron migration, however in my case the CPU instantly lost power after saving settings and exitting. The VID selected was 1.200 with VID special control of 133%, which I have used previously with other BIOS's to achieve a 1.56vcore BIOS reading, far below the maximum tolerance of such A64 chips, especially when temperature was well under control. I have also observed that there is literally no electricity running through the CPU (CPU is generating no heat at all, motherboard showing 4 diagnostic LEDs, usually indicating CPU is not even being detected), and to me this sounds like a blown trace or maybe a blown surface-mounted component - which usually would not occur unless a suddenly excessive high voltage is applied. Therefore, I believe the BIOS caused the motherboard to suddenly apply the max vcore that can be supplied immediately after I saved settings and exitted BIOS. I am concerned that other people may be affected, and so I think it's important that Oskar Wu, or whoever is the head of the DFI BIOS development team to be notified of the possibility of the presence of a bug in this BIOS. Again I have to emphasize that I cannot be 100% positive that the BIOS is at fault, but logically it seems to be the case.
Therefore, I speculate that the motherboard itself may still be functional - however I will not know until my new CPUs arrive. But as you would understand, I'm feeling VERY apprehensive and worried about testing my new CPUs on a board that is potentially dangerous. I do not want any more CPUs to be possibly destroyed by the board, even though I have resetted the CMOS - I still cannot feel secure about this. And yet on the other hand, if I apply for RMA with the board and they say there is no problem with it, then it would have been a big waste of time, and also money.
What is your suggestion to be the best solution to my situation? And, typically, under such circumstances, would AMD's RMA system cover the death of my Venice 3000+ CPU if it was caused by the motherboard? I understand that overclocking is not covered by CPU manufacturers, but in such a case, the death is most likely (as I stated before, I cannot be 100% positive) not directly to do with overclocking, but motherboard/BIOS related. Indeed, as you said, this has caused a large inconvenience for me, not only because I have go to through this RMA process, but mainly because I have lost a truly special CPU.
Any constructive help would be greatly appreciated, thankyou very much for your consideration thus far.
Kindest Regards,
John.
__________________
Any help would be GREATLY appreciated guys!
groove, fwiw ... I found that messing around with the various vcore settings can be very touchy ... I also work for low vcore with high multiplier, but that 1.20/133% was a no poster for me. try to stay in the 1.30 to 1.35 ranges with various multipliers..
however, a vcore conversation here is the wrong place ... this is a very import thread to many and we should try to keep this focused on vtt/vdimm issues at this point.
you should post support issues to dfi-street or start a different thread at this site.
just my 2cents
Didn't read the long post about the vcore. Anyways, for better vcore control, just do this vcore mod.
I have seen the same thing, it looks like a voltmod that has gone bad but when you reboot its back to normal.Quote:
Originally Posted by alpha0ne
But this was with an older bios and an untouched X850 card.
But you get eyes big as an octopus and heart stops beating when you see it.
As far as I know , this is beta bios ... And anything that screw up things could be happened ... You should not use it if you are ok with previous bios ...Quote:
Originally Posted by groovetek
yes, well, previous bios's had issues with async overclocking with my G.Skill UTT BH5 GH kit.
it looks like at dfi-street, someone has killed their FX55 sandiego the exact same way with this BIOS... maybe it's coincidence that it's with a 704 BIOS as well, but me thinks otherwise.
IMHO the err 'official' 310 bios is a 'kin disgrace :rolleyes:Quote:
Originally Posted by suio
i think the weird graphics corruption thing could be indirectly related to more voltage controlling bugs that could potentially be present with this BIOS.
I don't see any information that 7/04 bios solve the async issue ... So before you update to a bios that is still beta , you should know that could be happened , especially when there is unknow amount of ppl does not have any problem with the beta bios ...Quote:
Originally Posted by groovetek
Maybe here is the reason why...Quote:
Originally Posted by groovetek
http://www.dfi-street.com/forum/show...20&postcount=1
FARRRK me Oscar should have posted that HERE 2 days ago, not everyone relies on DFI street as the font of knowledge, MANY more do HERE :(Quote:
Originally Posted by tictac
Give Oskar a break... I seriously doubt he knew it two days ago... I'm sure he posted on "the street" as soon as he found it... and word got from "the street" to hear in a matter of hours (look in AMD section).
Oskar is by far the most responsive MB company engineer around and has busted his chops for the enthusiast community for quite a number of years. Sheesh....
/rant off
Peace :toast:
Not to mention you always take a chance when flashing a beta bios.. Im sure Oscar will have it sorted out soon..
Yep... Beta, just like Alpha BIOS should be flashed ***ONLY*** if one has issues with previous, Official BIOS.
Unfortunately, this might mean that even though BIOS glitch might have killed your chip, you're probably out of luck. Unless, DFI will be certain that that was the reason for it's death, then it would not really kill them to get you a new chip... It would be pretty good PR to me... Showing how much they care for users "beta-testing" their BIOSes.
Anyway, just take it easy. Don't attack Oscar Wu for that... get in touch with them one more time and see if they can help you with possible replacement or sth...
I haven't and have no intention of attacking him or even DFI at all - In fact I admire Oskar Wu for his honesty and willingness/commitment to solve the problem. I have, however, tried my best to warn other DFI nf4 users of the bug in the BIOS. I've been very rational with my emails and posts as you can see here on XS, and also on dfi-street, i4memory.
That said, I am confused how a BIOS could introduce such a fatal bug though, even if it is an alpha... I don't know how to code/program BIOS's, but assuming the code for VID doesn't change from 1 bios to another (since that's not where improvements are required/or being changed), it's surprising that this can occur...
I only used the 704 BIOS because every BIOS before that was giving me troubles with G.Skill GH BH UTT kit when using async memory divider settings. I had followed every guide, tried dozens of different settings, all with no proper stability. The 704 finally did it for me and I was very happy, but then unfortunately, this happened, when deciding to up my CPU speed (i need 1.56v (1.20 * 133%, tested in prevoius BIOS's) for 3.0ghz, it was set to 1.525v and no "Above VID %" setting prior to this)...
As for getting me a new chip - it would be good PR for DFI, but unfortunately it wouldn't happen with a large-scale manufacturer like DFI because they are already popular enough as it is... and where would they find me another week 17 chip that goes as hard as mine did, haha... I'd love to find another one that could though... sigh... :( not to mention as a uni student with no proper job atm, the cost of computer hardware really adds up. I had waited 3 months to build this system after selling my old parts because of exams and because I knew winchesters and AGP platform-related parts would lose value very soon after that, so sacrificed all this time so I could gather enough money for these parts, and then on the 3rd day of setting it up it dies on me, very very bad luck :slapass: .
edit: if anything though - it seems at least here in Australia, AMD's RMA system is extremely stringent, and I may not be able to get a replacement under these circumstances... if that is the case, then it'd be good if i did get just any 3000+ replaced for me... although do undrestand that DFI is not obliged to do this, since it was an alpha BIOS...
[on topic]
i did the vdimm mod the other day but located the pot down neat the chipset. this means i have the wires about 6inches long that goto pin 4 and 6. now that i think about it though, is that a bad idea? wouldnt i be introducing a lot of EMI and loss?
thanks
[/on topic]
You can reduce some EMI by wrapping the GND wire around the Hot wire for as far as you can until you get to the solder points. You could glue the pot to the FET Heatsink thats right next to LM358 if you want to feel safer.
-CaT
So i have had a few mails asking about the mods I did. i know Malves has already posted but these still may help someone for the Vdimm. i have added the VTT also with the connections explained.
VTT:
http://homepage.ntlworld.com/a.leach...mm/VTT-DFI.JPG
Vdimm:
http://homepage.ntlworld.com/a.leach.../Vdimm-dfi.JPG
VTT is worth it IMO, i was able to gain a few fsb at 2.65V 300fsb+ and it did seem more stable, the pot swings the voltage either way.
any suggestions on how to tweak these two and there interactions? For example, I assume set the vdimm mod first, and then tweak the vtt mod to be 1/2 of vdimm?
Is that the idea?
any measuring points on the vtt mod better than others??
Thanks!!
measure VTT on the 9173A, I just set Vdimm slowly and tweak VTT along with it. The dip is still there, but I was able to tweak VTT a little high, this reduced ther dip and gained stability.Quote:
Originally Posted by timpanogos
next mod is to add another 9173 and see if the dip is reduced even further...all i have to do is find a source for the chip.
I was supprised to see how little Vdimm is actually really needed, 2.75V or so on a few dimms and usually 2.65 to 2.68V max's the dimms out.
hey bigtoe, thanks, very interested in the VTT mod. How much better is the VTT fluctuations with only this mod ? Or should we wait 'til we have some more info regarding a mod with maybe another 9173A ? I really want to totally eliminate those VTT fluctuations.
Regarding the mod: might it be possible that, instead of soldering to an R there is a pin right on the DIMM slot, some PIN where we can do that instead of the need to solder around on that resistor ? Just wondering
--- Flexy --- Pin 1 & Pin 7 of the DIMM slots (Vref & Vdd respectively)... will go right down into the test point clips of the DIMM socket... this would allow you to raise Vtt only, no lowering (which you don't really want anyway). From my previous testing, doesn't change magnitude of the dips, only offsets them so absolute value isn't so low. Or use 1, 93, & 96 for up/down control (Vref, Gnd, VddQ respectively). Center leg of POT to pin 1 (Vref), outside leg(s) to other pin(s).
--- Tony --- Only place I've found any RT9173As was a distributor in NY stateside... and they had a minimum order of 100 :/ Just rob one off one of your other MBs :p:
ouch on the min order!
I have two questions for you folks :)
My 50K VRs (I ran out :() should be here tomorrow, I'm going to install VTT and VDIMM voltmods on my board when they do get here.
~ How hot can VTT be run,without compromising stability?
~ Is a hot VTT better than a low VTT for stability?
~ At what point does an above-reference VTT begin to adversely affect stability?
~ When you say "10K OHm VR set to midpoint", you mean 5K of resistance right?
Thanks, I'll be sure to share my results!
I just found something which brings all sorts of ridiculous ideas to mind, a 2 farad capacitor pack that came out of a dead power amp. It's like the size of a car battery, but if it has the potential to fix these VTT problems I would love to waste it on a motherboard. What points should I try attaching it to? (No rush, I'm out a videocard right now because of neweggs crappy refurb sales, at least they have a good RMA service)
Check the quotesQuote:
Originally Posted by felinusz
Can't do that m8, all the boards I have here are different steppings, i need them for support.Quote:
Just rob one off one of your other MBs
I have been looking over the dead boards and none have the IC...looks like I will have to just find someone willing to sample them.
If someone can design a simple circuit for fixing the VTT flux, that could easily and simply be attached to the DFI nForce4 motherboard via pomona grabbers, I am sure we could easily find 100 people interested in a fix that cost ~$5-$10 to put together - practically every A64 user has this motherboard.Quote:
Only place I've found any RT9173As was a distributor in NY stateside... and they had a minimum order of 100 :/ Just rob one off one of your other MBs
Of course, there is the issue of testing the circuit out first, which is impossible without one of these ICs in hand...
Thank you Bigtoe, my board goes on the workbench right now (my 50K VRs just got here, mouser pulls through again ;)), I'll have some results from my board by tonight or tommorrow.Quote:
Check the quotes
I just need to play some HL2 first (I always play a good hour or two of whatever game I'm enjoying before doing any voltmods - that way if I kill something I'll at least have had a good run of fun before doing so ;))...
yes....but then i really dont know whether 100 people would actually know what a VTT is nor ever heard of that there are issues with it.Quote:
Originally Posted by felinusz
Anyway i would buy/do that mod in a heartbeat...you could defintly count ME in...so we only need 99 more :)
... 98 ...
→97
--- Craig --- Don't do it :) Two problems... first is in-rush current, second is effective ESR and ESL. First one might fry your board, second one would make your mega-cap ineffective.
>>> teaser <<<
http://xtremesystems.org/forums/atta...id=34041&stc=1 :p:
You're kidding, right? :rolleyes:Quote:
Originally Posted by flexy
Malves, Bigtoe; your mods are done and work perfectly - thanks a lot guys!
The VDIMM mod did not get me any more or less stability when at the same overvoltage - before the mod my BH-5 ran at 278 MHz with complete memtest86 stability at 3.78V, after the mod it still peaks at the same frequency with that overvolt.
However, the mod looks to have netted me another 1 MHz (so far, I have only done very brief testing!) on my stable overclock, due to the VDIMM overvoltage fine-tuning that it allows. My board overvolts VDIMM a great deal, the overvolt increasing as BIOS VDIMM increased. At 3.7V in the BIOS, the board gave me 3.78V. At 3.8V in the BIOS, the board gave me 3.9V - all the overvoltage values in between this 0.12V gap were unnattainable. Now at 3.83V, my memory is running at 279 MHz with stability (for the ~10 minutes I have run it through memtest T5 anyhow), and I expect the peak bencheable speed shall have increased by ~1 Mhz as well. I have been running my BH-5 at 3.78V day-to-day... I expect this mod will be very useful.
The VTT mod works perfectly, and I took the liberty of making some multimeter measurements. In memtest86+ test#5, VTT is stable as a rock, as is VDIMM. In memtest86+ test #8, I see the same thing all of you are seeing - my VTT fluctuates between 1.91V and 1.94V (with VDIMM set to 3.83V). The VTT mod is a decent fix, I tweaked VTT to run a little hotter, so that it does not droop below reference - and the hot VTT does not adversely affect stability (although from my extremely brief testing thus far, it does not appear to be beneficial to stability either).
I'm going to spend some more time with the improved voltage control tommorrow, and see if more VTT fine-tuning will take this memory any farther.
96.Quote:
→97
95 :) , and I think flexy you underestimate how many ppl really do know about these things :toast:
Hrmm..were working on a circuit for the VTT fix..it will have to be something that is purchased and then soldered on or clipped on with leads, its not just a matter of soldering a VR in place :P
-CaT
Does anyone think my attatched circuit would work? R4 is used to decrease Vtt that the RT9173A gives out, so that only the new regulator controls Vtt, and R2/R3 are feedback resistors to provide gain to compensate for R4 to bring Vtt back to 1/2 Vdimm.
Does anyone know the value of the voltage divider resistors for the Vtt suppy so that I can put in some values for the resistors?
I'm using a similar circuit to this on my NF7-s to privide a Vtt supply that follows Vdimm (it doesn't have R3 or R4 and R2 is just a link to make it into a unity gain buffer). The IRF630 is rated to 6A iirc, so should be able to handle the Vtt supply.
Well, yeah, the VTT mod that bigtoe posted isn't a fix for the fluctuation, merely a means of running VTT slightly hot in order to minimize the negative effects of VTT dropping below reference 8-/.Quote:
Hrmm..were working on a circuit for the VTT fix..it will have to be something that is purchased and then soldered on or clipped on with leads, its not just a matter of soldering a VR in place :P
VTT is still all over the place under certain conditions, and it's bad enough that it's even visible with a cheap multimeter.
I sent Richtek a request for samples, will see what happens.
Simplest solution would be piggybacking a second 9137A. Use a piece of copper on TAB to raise height, use some AS Ceramique on body, and bend @2's pins down to solder onto #1's.
Its not a fix, but it does help, i have a feeling by the time we will see a true fix there will be a new board out without this issue ;-)Quote:
Originally Posted by CaTalyst.X
*the* new board... the one with the ati gruper PWM by chance? lol :p:Quote:
Originally Posted by bigtoe
just because there is a new board out doesnt mean everyone will upgrade. so the vtt fix will help people for a long time. even those that havent upgraded from socket A yet becuase of their low budget...Quote:
Originally Posted by bigtoe
Quote:
Originally Posted by stealth17
I agree.. Im sure there will be plenty of us including me doing the mod once its complete..
--- Cat --- A replacement regulator is indeed best way... but cost is higher ;)
--- Persivore --- Pin 4 of the RT9173A IS connected to the midpoint of the divider... so in effect you have pin 2&3 shorted together on your schematic if you leave Pin 4 of the RT9173A connected to the MB. ;) You would need to either lift pin 4 on the RT9173A (and then ground it to disable the RT9173A's output), or simply lift pin 5 which is it's output so it's out of the circuit. The resistors of the VMem divider used to generate Vref are 10 ohms. 2nd comment on your circuit is that FET will be seeing about 6W average and 12W peak using +5V as the source. 3rd comment is that a Vtt regulator must be able to not only source current, but also sink current to function properly ;)
--- uwackme --- Richtek is gonna wonder why the interest from so many places ;) BTW... I was thinking something more like the first attachment below for mounting method... remember the tab is connected to Vmem.
-----
The 2nd attachment is a block diagram of one of the 3 alternate methods I previously mentioned as a possibility (other 2 already discussed)... since no real power in the circuit, no HS required...could be made on a small perf board. Basis for it is to adjust the reference input to the RT9173A so it's output tracks properly under load, increasing it's reference when it starts to droop. Caveat - the RT9173A can't be going into current limiting for it to work (which it may be doing).
Peace :toast:
I like the idea EMC, good solution to mounting. Only hard solder is the Vmem to #2 regulator...dont want to fry it, but the conductor needs to be 18GA or better.
Idea is to solder #2's pins directly to #1's pins... no wires ;) (the tab is VCtrl input - it isn't connected to the tab for current capability - DFI just used Vmem to supply Vctrl)
With Vctrl being Vmem, maybe we are seeing a amplification/feedback of slight noise/load-droop in Vmem back around into Vtt? Wonder if pulling Vctrl in the air and attaching to 3.3V or 5V rail would magically make the droop vanish?
I would use 3.3V but I have a feeling it would help.Quote:
Originally Posted by uwackme
hmm maybe i do :) On the other hand i can see many enthusiasts having especially the dfi nf4 :)Quote:
Originally Posted by alpha0ne
No, doesn't vanish, but helps lower RdsOn and thus minor improvement in droop... Oskar already tested that in this post. Haven't heard back if he's gone to the 4V+ level yet.... if it wasn't such a pita to do, would have already tried 5V myself ;)Quote:
Originally Posted by uwackme
Peace :toast:
i know this is about the mod, but could anyone give information how the OCZ booster would affect the VTT issue ? I still did not hear conclusive things regarding the booster and the VTT stability.
If it works it would be a alternative for people who dont want to take risks soldering/modding on the board and it would be relatively cheap to do.
Does the booster even interface with Vtt? Isnt Vtt only supplied to the terminating resistors, and not to the DIMM socket at all?
Wonder if shorting Vref to Vtt and using a booster would then let the Booster's regulator Vref output help stabilize Vtt?
Vref has a seperate supply correct? There is no droop on it.
Looks like the ultimate solution is for DFI to come out with a new rev of MB and we go buy that.
That might be the ultimate fix, but for everyone else that got the board that isn't fixed (if it even is or will be) then finding a mod that solves the problem is a must...and I'm sure with the people here we can do itQuote:
Originally Posted by uwackme
Short Vref to Vtt and you'll have an oscillator and maybe toast :p:Quote:
Originally Posted by uwackme
Ugh, didnt realize Vref was just off a resistor voltage divider. Yuck.
Sucks that Vctrl is on the tab, tie it to 3.3Vrail, setup a seperate POT based voltage divider to dial in 1/2 Vdimm value'd Vtt and see if droop vanishes.
What kind of bypass caps have they implemented on the Vtt resistors? Starting to think they didnt use ANY, where there should be 560pf and .1uf ceramic cap for every 4-8 resistors in the terminator network.
Did you see the "rev2" board layout.... looks like the same Vdimm/Vtt circuit.
Sigh.
At this point it begs the question.... any other NF4 boards having such issues? Time to consider Asus or Abit and just voltmod them.
hmmm....any news ? My finger is itching and wants to mod...this issue is too serious to just ignore.....if i knew the OCZ booster helps i would get one....but still no real information if it would affect VTT..
What about a voltage divider between VIN and REFEN?
-CaT
--- Flexy ---
Waiting on some parts... something about a slow boat... may be a bit :/
seems the vdimm mod made my vdimm unstable. jumps a lot now. anyone else have this?
As far as I know, the DDR Booster does not directly fix VTT tracking issues.Quote:
hmmm....any news ? My finger is itching and wants to mod...this issue is too serious to just ignore.....if i knew the OCZ booster helps i would get one....but still no real information if it would affect VTT..
Also, you don't want to use a DDR Booster with high voltage memory and a VDIMM >~3.65V, as the Booster will begin to fluctuate VDIMM a great deal between idle/load above that range. For ~3.8V under load I needed a good ~4V at boot with a Booster, and the loaded voltage fluctuated.
Well? Any ideas? Im still waiting for the sample regulators to arrive.
EMC....
Been thinling about this.
I really think the key to the droop is the fact that Vref is a simple voltage divider, 10ohms/10ohms and then this is used as the reference for the 9173.
The instantaneous current draw on Vref could cuase a slight dip on Vref that is being seriously amplified by the 9173 and is seen amplified on the Vtt output.
Riddle me this batman. What if we instead used 10K/10K resistors on the dividor, and cut the Vref supply connection from the divider-junction and instead tied it to the Vtt output of the 9173?
Vref = Vtt supposedly anyway, and they are driven from the same point on boards like the NF7-S, so it is no something forbidden.
That way there is NO current component to the voltage divider (our reference point for generating 1/2Vdimm) so it should be totally stable. And Vref/Vtt get plenty of juice and I think the "droop" will vanish.
I think the divider and how it is being used was a nice $$-cutting trick, but it is introducing a dip that gets amplified to a droop by the 9173. A way to see if the divider is involved at all is to put a .01uf cap between the divider Vref point and ground...right at the resistor junction. Your measured "droop" should be reduced somewhat.
Thoughts?
I would suggest you go back, look closely at the scope captures in posts #62 & #75, and rethink bro ;) One of them also answers your pondering regarding 256x2 versus 512x2 in that other thread...Quote:
Originally Posted by uwackme
Other thoughts:
- Vref fluctuating with Vtt would tend to make matters worse, not better. It isn't a reciprocal relationship.
- A good half dozen filtering options on all three inputs to the RT9173A have already been through the dungeon - net result zero.
- Temp has a larger effect than filtering, but it does not solve the problem, only lessens it slightly (the RT9173A being next to the Vmem FETs and sharing the copper Vmem plane doesn't help with this one when using BH/CH memory either)
Have decided that between the slow boat, part availability issues, and my present (excessive) work load and other commitments to dump option A and have gen'd up something that'll allow options B or C -- fixing the RT9173A's tracking under load (if it can be and isn't a pure OC throttle problem) or completely replacing it if it can't. But it's something that isn't wired in easily without a PWB due to packaging (25mil spacing) so one is in process. ETA 2-3 weeks.
BTW, based on currently available pics of the "next rev" DFI board, they still are using the same VTT regulator :rolleyes:
Good luck on your boat shipment and let us know how it works out when it arrives bro.
Peace :toast:
Welp i did the mod today at tonys urging..
here are some pics of my handywork.. i jacked the jump idea from tony as it allows for easy removal of the mod..
I have yet to mess with it, but i hope it helps out.
Nice job. Did you gain stablity after the mod? How's the vdimm working out?Quote:
Originally Posted by MaxxxRacer
From the earlier posts, I understand that this mod will shift the vtt "profile" up a little, reducing the effect of the downward swings, right?Quote:
Originally Posted by MaxxxRacer
But what types of users will this help? Those using TCCD? BH-5? Only overclockers? One poster in this thread said his experience was that over time he needed to increase vdimm on the TCCD from 2.8 to 3.1 over time. Will this mod help with issues like that?
Well both my TCCD and UTT is borked right now so I reallly cant comment on either one being more stable, but it does allow for VERY fine adjustment of hte VTT and VDIMM. I used a 20k pot on the VTT becasue i iddnt have a 10k, and it allows me .0001 adjustment resolution, which is even further than my multimeter shows.
As to the vtt, I have it boosted about .01 volts or so. when i get my UTT and TCCD back from RMA I will let you guys know how this mod works out.