HELP - 9600GT VoltMod Problem

[dadi_oh]

So I decided to try the voltmod on my 9600GT following the instructions here on the forum.

http://www.techpowerup.com/articles/overclocking/vidcard/150

I was shooting for the 1.3V mod and successfully removed vid4. Next I needed to short out vid5 to complete the mod. My method has been to use a fine strand of wire to bridge the two points. I keep the wire long, bend the end at a 90deg angle and solder across the point. Then I just snip off the long end.

I got the wire soldered across the two points and unfortunately as I was trying to snip it I tugged on the wire and it pulled off the trace from the right side of vid5. :eek:

So now, what I need to know is where does that right side of vid5 get connected to? There was a bit of the trace left and I "think" I managed to get the tip of my meter onto it and I "think" it is connected to the point highlighted in my attached picture.

Can anyone verify this for me? If that is the point it connects to then I will probably put a wire between the 2 points to achieve the connection without the use of the trace.

Yes I know I'm a bonehead. No need to point it out. These are the risks of soldering something smaller than a gnats nuts.

[Solaris17]

you want ot circle the race you removed?

[dadi_oh]

you want ot circle the race you removed?

Sure thing. See below.

[Solaris17]

wait so you melted through the pcb and severed the connection?

[dadi_oh]

wait so you melted through the pcb and severed the connection?

Nope. The wire was soldered to the pad that the resistor would normally attach to. When I tugged the wire it pulled the pad off the surface of the board and lifted off the trace (which is run on the surface layer). There was a little stub of the trace left so I managed to use an exacto knife to scrape off the solder resist to expose the copper. I managed to get my meter probe onto that exposed copper (I think) which is why I "think" it is attached to the point I indicated. But before I do that I wanted to be certain that is the correct connection point.

[Solaris17]

just draw a diagram? im sorry im tired and im not sure what your talking about..mainly because i wouldnt know what you were doinf soldering all the way over their but its fine im not raggin ya iv done some dumb shit but i see no problem just dropping a ball of solder on it if thats the case no need for wires.

[dadi_oh]

just draw a diagram? im sorry im tired and im not sure what your talking about..mainly because i wouldnt know what you were doinf soldering all the way over their but its fine im not raggin ya iv done some dumb shit but i see no problem just dropping a ball of solder on it if thats the case no need for wires.

I'm probably not explaining it well. I work in electronics so some of my terminology might sound strange. The trace that I circled in my second diagram runs under a component on the board (I assume it is a transistor). The trace is on the surface layer of the board and covered by something called solder resist (usually green in color if you are Nvidia but it might be red if it was an ATI board :) That trace is a very thin piece of flat copper wire that can not take a lot of heat (like in the soldering process). Traces do not connect directly to component leads for this reason. They first go to what is often called a "pad" which is what the end of the resistor is soldered to. It is a rectangular piece of copper that matches up to the solder lead of the component.

What happened in my case was that the heat of the soldering process probably delaminated the trace from the surface of the board. Then the inadvertent tug on the wire pulled the pad off the board and part of the trace with it. This left a small exposed section of the trace still under the solder resist. I scraped off the solder resist to expose the shiny copper of the trace and that is what I was probing with my meter.

Now the pad is completely gone so I can no longer solder a resistor (or piece of wire to it). There is just a smouldering hole in the FR4 (the material that they make circuit boards out of).

Luckily for me, the default voltage does not have this resistor populated so I am still running at stock voltage. If I had messed up one of the others I might have a non-functioning card.

Hope that makes more sense. A picture might help but the card is back in the machine with it's new waterblock attached so I may not pull it out for a long time. Unless the lure of higher voltage and breaking the 800MHz core lures me there :laugh:

Time to get some sleep :toast:

[dadi_oh]

OK. I found this other diagram on another site. This is for the reference 9600GT design. It lines up with the guide here at TechPowerUp. Note that a "0" means shorted (with zero ohm resistor) and "1" means open circuit.

This would suggest that the 2 points that I labelled with yellow arrors would be connected to each other. Therefore if I were to connect the points indicated by the "magenta" line then I would achieve the vid5 connection required for volts above default.

Some of the comments on the other site suggested that OVP protection was still an issue above 1.2V but the guide here at TechPowerUp says that OVP is over-ridden with this mod. Anyone try the mod to 1.3V or 1.4V and can confirm that OVP doesn't kick in? I would hate to go through the mod just to discover that I am bound by OVP.

The piece of the puzzle I am still missing is whether, indeed the points indicated by the yellow arrows are connected to each other. I can't measure it on my board because the pad is torn off. Anyone care to measure it on their board. Please? Pretty Please? :)

[Solaris17]

i dont see in any way hnow that tells me what you removed...thats a vmod guide....

[dadi_oh]

i dont see in any way hnow that tells me what you removed...thats a vmod guide....

OK. Following the principle that a picture is worth a thousand words here are two pictures (so I guess that is 2000 words worth).

The first picture is what this area of the board looked like before I tried shorting out vid5.

The second picture shows in RED the trace and pad that are torn off. i.e. where you see the shaded red area there is now no copper.

[Skitzo]

well from what i can tell the trace leaves the board and goes up to the component underneath. sorry for the poor pic but this camera sucks. Can you read the part number of the chip, mine is hard to read. If i read it right it was RKS02; I couldn't find a data sheet or description but hey it could be hanging out with the datasheet for adp3208 :)


I'll hunt for my mag glass, that should help...

[Solaris17]

ooooooooooooooooooooooooooooooo ok so the second pad on vid 5 is gone?....thats fone if you dont want them connected well you did the job and if you do want them connected just drop some solder on both if your worried the trace isnt connecting to the pad just dable a mico ball back on the copper youll be fine you didnt seriously break it just make sure you take your time :)

[dadi_oh]

well from what i can tell the trace leaves the board and goes up to the component underneath. sorry for the poor pic but this camera sucks. Can you read the part number of the chip, mine is hard to read. If i read it right it was RKS02; I couldn't find a data sheet or description but hey it could be hanging out with the datasheet for adp3208 :)


I'll hunt for my mag glass, that should help...

Cool. So the 2 points that I think are connected are circled in yellow. Do you have an ohmeter that you could buzz out those two points with?

Thanks man.

[dadi_oh]

Cool. So the 2 points that I think are connected are circled in yellow. Do you have an ohmeter that you could buzz out those two points with?

Thanks man.

Or alternately these 2 points might also be connected. If you have an ohmeter and can measure it for 0 ohms that would be awesome. That would be an easier solder job for me... not so cramped.

[Skitzo]

the lead looks like it connects up into rks02 between the two big contact points... wish i had a better camera... sorry

I can only find my swiss army knife mag so it's really hard to see clearly, but to me it looks like a silver wire not a trace where it goes up and i cannot see the trace continue past that.

I tried to find a data sheet to check the number of connections but no luck.

[Skitzo]

I dont't have time at the moment, I have to leave for a few hours but I'll see what I can do later...:banghead:

[dadi_oh]

ooooooooooooooooooooooooooooooo ok so the second pad on vid 5 is gone?....thats fone if you dont want them connected well you did the job and if you do want them connected just drop some solder on both if your worried the trace isnt connecting to the pad just dable a mico ball back on the copper youll be fine you didnt seriously break it just make sure you take your time :)

Yup. The pad and trace tore right off the board :banghead: Unfortunately there is not enough of a scrap of the trace left to solder to. That's why I am trying to find an alternate place on the board that the trace was connected to. I believe that the pad opposite the pad that I tore off is the ground. So I am trying to find the other end of the trace to attach to ground and achieve a short circuit for vid5. All voltages above stock require vid5 to be shorted. If I leave it open I am stuck at stock voltage.

[Skitzo]

if it was a ground wouldn't any ground work? lol sorry can't read i guess n/m

[dadi_oh]

I dont't have time at the moment, I have to leave for a few hours but I'll see what I can do later...:banghead:

Hey no problem. The card is humming along fine at stock volts. I need to screw up enough courage (and a better, finer soldering iron tip) to try it again. A little voice in the back of my head is saying "doooooonnnn't doooooo iiiiit.... yoooou'll beeeeee soooooorrry". But I find the voice changes once I have a few beers into "Yooooou daaaaa maaaaan. goooooo for iiiiiit. Whaaaat dooo yoooou haaaave tooo loooose?" :laugh:

Unfortunately the combination of not listening to the voice and consuming beer leads to poor soldering.

[dadi_oh]

if it was a ground wouldn't any ground work? lol sorry can't read i guess n/m

No, I meant that the trace that tore off was meant to connect to the ground point in order to short the vid5 circuit. But maybe it isn't a ground. I would need to pull it out of the case and do some more ohmeter measurements. If it were ground then you are right... any other ground would do. If it isn't ground then I need to run the wire back to that point specifically. Either way. The length of the wire doesn't matter. Just a good soldering job at either end ;)

[Skitzo]

is there a way to relaminate the pad to the board? Is there a contact with that particular pad to anything else? If not, why couldn't a person try to use a pad or something similar, make the solder conection that is needed, relaminate (glue) the pad by whatever means it takes to make it stck and repair the trace connection with conductive ink/paint.

I'm not sure what it would require to attach the pad so this mat not be feasable.

if the mod is only using solder connections and not 0 ohm resitors why not connect the trace to the complimentary pad that hasn't been removed by means of conductive ink. Skip the damaged pad all together.

Again these are just ideas so put alot of thought into anythng I may suggest before hand, should you decide to try it.

[TheGoat Eater]

Tip #1 - when in doubt ask largon ;)

[Skitzo]

ask everyone, all info has value, maybe just not where you want to apply it.:)

seems like it would be better try a repair apposed to a work around, but that's just my view.:)

[dadi_oh]

is there a way to relaminate the pad to the board? Is there a contact with that particular pad to anything else? If not, why couldn't a person try to use a pad or something similar, make the solder conection that is needed, relaminate (glue) the pad by whatever means it takes to make it stck and repair the trace connection with conductive ink/paint.

I'm not sure what it would require to attach the pad so this mat not be feasable.

if the mod is only using solder connections and not 0 ohm resitors why not connect the trace to the complimentary pad that hasn't been removed by means of conductive ink. Skip the damaged pad all together.

Again these are just ideas so put alot of thought into anythng I may suggest before hand, should you decide to try it.

The trace is torn off right to the edge of the component that it runs under so trying to get a good solder joint onto it would be a challenge. I am "pretty" sure that there is nothing else attached to the pad that tore off since that would require a "via" next to the pad and there is not one there. So basically all I am trying to achieve is to find the other end of the trace that I tore off and just use a small wire to replace it.

The trick is to use a very fine solder tip and just a momentary contact to avoid too much heat on these small surface mounted components. Next time I will take it into work with me and use one of our soldering stations. They have binocular microscopes that make this sort of fine soldering much much easier. I was too impatient this time around and am paying the coinsequences.

The card is running happily at default voltage right now at 790 core and 2100 on memory so no particular rush at the moment.

Thanks for all your help. It is really appreciated.

[Skitzo]

Sounds like the perfect setup for modding, I'm feeling a little jealeous lol. I don't have access to that sort of setup so i tend to find solutions with the resources I have. people have been doing volt mods with conductive ink/paint for as long as can remember. It seems to work well, it has to be better that a pencil, but a proper solder connection would be best. Would there be enough conductivity to do a repair like this with conductive paint? A fairly small path could be created with a toothpick and the trace itself isn't very large so it has me curious. It should be possible to get a connection on the trace as long as there is bare metal.

[dadi_oh]

Or alternately these 2 points might also be connected. If you have an ohmeter and can measure it for 0 ohms that would be awesome. That would be an easier solder job for me... not so cramped.

Well I pulled my card out and decided to verify a few connection points. I was correct about the connection. The points indicated by the dotted red line are indeed connected together. So all I had to do to repair the card was to bridge the two points conncted by the pink line. I then connected the bridges to get 1.3V on the GPU core (rather than the default 1.1V).

I fired everything up and it booted to windows fine. I fired up ATITool which I use to check for artiacts. Basically I use RIVATuner to set the clocks and the artifact checker in ATITool to verify operation (too bad RIVA Tuner doesn't have an artifact checker... or maybe it does and I don't know about it...)


So at stock 700MHz everything was fine. I bumped up the core to 790MHz (a know stable core value before the mod) and after a few seconds spinning the fuzzball in ATITool the screen went blank. No recourse except to power down. This sounded like the overvoltage protection kicking in from what I have read. But I thought that this vid5, vid4, vid3, vid2 method was supposed to avoid overvoltage protection???

So I took the card back out and adjusted the vid bridges to give me 1.2V. I had read somewhere that people could stay below the overvoltage limit at 1.2V. That seemed to work better. I could get up to 840MHz on the core and ATITool was reporting no artifacts. At 850MHz I would get the blank screen. So with the core at 840MHz and memory at stock I fired up 3dMark06. Everything was fine until the second scene in Firefly forest where Twinkle-Dee and Twinkle-Dum are flying up the side of the hill and... boom... black screen. Doh! Had to power cycle again to get things back. But this time, part way into loading windows I got some screen corruption and the system reset. Had to power down, wait 2 minutes, and power on to get back to normal. (If I powered down and only waited about 10seconds the same corruption issue was still there... weird). Seems like something had to cool down???

To make a long story short I tried various cores (keeping mem at stock) and it was not until 790MHz that I could get 3dMark06 past that point int firefly forest. Funny thing was it always failed (blank screen) at exactly the same point in that bench. Even if I skipped the first proxycon (or whatever they call it) bench. Very predictable crash point but I have no idea why...

So 790MHz seemed OK. I put 3dMark06 into an infinite loop running the SM2.0 and SM3.0 tests. Wateched 2 cycles of it and went to bed. Got up in the morning and there was the black screen again.

So just to summarize:

Vcore V Freq ATITool 3dMark06
1.3V 700 OK
1.3V 790 crash
1.2V 700 OK
1.2V 790 OK
1.2V 850 crash
1.2V 840 OK black screen
1.2V 800 OK black screen
1.2V 790 OK OK for about 15 minutes. Crashed during the night.


Next step I guess is to attach a couple of wires to the Vcore monitoring points to see what my actual Vcore is and whether it shuts down (indicating overvoltage protection).

BTW during all of this my max temp on the GPU core was about 35C @ 1.3V and about 33C @ 1.2V so this doesn't seem to be a temperature issue.

Any other ideas?

[Skitzo]

howw is the cooling on the rest of the card, gpu is covered, got enough heatsinks where they need to be? need more airflow on existing heatsinks? This is why I would likw to build in a voltmeter, not always useful, but when it is it is... ocp perhaps, wonder if tha has been solved. I asked about the data sheet for the phase reg at a local industrial electronics store, buddy was stumped, couldn't figure out why it wasn't available.

[Skitzo]

not exactly sure how these vid mods function but, would it be possile to use a resistor other than 0 ohms to hit say 1.18? Or are the mods more of a trigger and the resistance is built in.

[dadi_oh]

howw is the cooling on the rest of the card, gpu is covered, got enough heatsinks where they need to be? need more airflow on existing heatsinks? This is why I would likw to build in a voltmeter, not always useful, but when it is it is... ocp perhaps, wonder if tha has been solved. I asked about the data sheet for the phase reg at a local industrial electronics store, buddy was stumped, couldn't figure out why it wasn't available.

When I removed the stock heatsink I made note of the components that had thermal transfer pads on the heatsink. Basically just the memory and the FETS in the power section. I put Zalman VGA heatsinks on the memory using 3M thermal tape (I thought about thermal epoxy... I have some... but if I ever want to resell I need to put the heatsink back on). Given the rather questionable thermal pad used on the memory in the stock heatsink I am thinking these cool at least as well, although that is a guess. I also had some RAM heatsinks that I placed on the 9 FETS. 1 heatsinks on each group of 3 for a total of 3 power heatsinks.. These also are not that warm to the touch. In my case I have a 120mm fan at the front, directly in front of the hard drives. The hard drives are spaced apart to allow a nice clear channel back to the video card so I think there is decent airflow over these heatsinks. I also thought briefly about mounting an 80mm fan on standoffs on the bottom of the case pointed up directly at the video card. That would give a direct blast of air to these things but doesn't do much for the nice front to back airflow I currently have setup in the case.

I'm not convinced this is a thermal issue. The other thing that it could be is that my Antec 400W power supply is not up to the task. I read some threads that questioned whether 400W was adequate. I originally had a 650W SLI supply on this system but moved it over to my son's setup since he has SLI. I have a spare Antec 450W supply that has a couple of extra amps on the 12V, 3.3V, and 5V rails. I could easily switch that in here and see if the extra power helps.

[dadi_oh]

not exactly sure how these vid mods function but, would it be possile to use a resistor other than 0 ohms to hit say 1.18? Or are the mods more of a trigger and the resistance is built in.

No, these are zero ohm resistors to just provide a logical low level to a digital input. Basically they feed a binary word into the input of the voltage regulating circuit. So the presence or absence of a resistor is just a binary "0" or "1" to the voltage circuit rather than an analog voltage. That is the table that i posted in one of my earlier posts.

[dadi_oh]

The more I think about this the more I am thinking it could be a power supply issue. Not sure how much more juice the 9600GT pulls on the 12V rails going from 1.1V to 1.2V but I wonder if that was enough to put me over the edge. My system is fairly lightly loaded... 2 hard drives, 1 DVD drive, the 9600GT, P5K-E motherboard with E2160 @ 3.3GHz. Whenever I have looked at the voltage rails with a monitoring program (like ASUS PC Probe) they seem to be fine.

Come to think of it, the CPUZ guys over at CPUID wrote a HW monitoring program that stores max and min values for a bunch of things, including voltages. Maybe I should keep an eye on that and see if one of the rails is dropping just before the crashes... hmmm... that could be it.

That might also explain the weird issue where I had to power off the system and wait before powering on to get it to boot properly into Windoze. Maybe something overheated in the power supply. I could boot into BIOS after a crash and see if one of the rails is running low.

[Skitzo]

sorry, yeah should have looked there first. General opinion seems to be 500W psu with a 9600gt. I think they were listed to work with 400w in the begining but that seems to be in debate. any idea what you liquid cooling system draws?

[dadi_oh]

sorry, yeah should have looked there first. General opinion seems to be 500W psu with a 9600gt. I think they were listed to work with 400w in the begining but that seems to be in debate. any idea what you liquid cooling system draws?

Good point. Not sure how much the pump draws. And for that matter I do have 5 fans on this system all running on 12V. 2 X 120mm Antecs, 2 X 120mm on the swiftech rad, and 1 ASUS NB fan. That might add up. I can't find a current consumption listed on the Antec site.

I have a clip-on current meter. I think I may try checking how much current is being drawn by the various components. Maybe the 450W isn't going to be enough either. Maybe I am in the market for a power supply... Recommendations?

[Skitzo]

i picked up a corsair 520W modular when i added the 9600gt, I'm hoping this will cover my liquid cooling needs as well. I have kicked myself a couple of times already for not going bigger, too much is better than not enough. gotta try to take into consideration any future upgrades.

[dadi_oh]

Good point. Not sure how much the pump draws. And for that matter I do have 5 fans on this system all running on 12V. 2 X 120mm Antecs, 2 X 120mm on the swiftech rad, and 1 ASUS NB fan. That might add up. I can't find a current consumption listed on the Antec site.

I have a clip-on current meter. I think I may try checking how much current is being drawn by the various components. Maybe the 450W isn't going to be enough either. Maybe I am in the market for a power supply... Recommendations?

The plot thickens... or is that "thins"? I think I am getting closer to solving this.

I went out and bought a new OCZ Stealth Stream 600W supply which rules out any power supply issues.

I decided to monitor what was happening with my actual Vcore on the GPU and discovered something interesting. Right now I have: vid5 - shorted vid4 - open vid3 - open vid2 - shorted. According to the chart this should give me 1.2000V on the core. I grabbed my trusty Fluke meter and measured the actual voltage and got the following:

idle = 1.2100 (i.e. 10mV above spec)

Not a big deal but here is the interesting part. I get different load voltages depending on what benchmark I run and also what core voltage I have set. For example, when core is set to default 700MHz and Proxycon benchmark is running the Vcore rises to between 1.234V and 1.240V. i.e. goes up by about 25-30mV or so.

Now remember in my past post I was saying it always crashed at the same point in Firefly Forest? When it hits that exact location the Vcore peaks at 1.246V. It did not crash however since I am only running at 700MHz core. Both Canyon Flight and Deep Freeze are similar to Proxycon, peaking at about 1.240V. So Firefly Forest draws more power for some reason:confused:

So next I set the core frequency to 840MHz and ran the benchmark. Well this time Proxycon hit about 1.244V peak (or about the same as Firefly at 700MHz core). Once it finished Proxycon and started Firefly Forest the voltage jumped up to about 1.248V. At the exact point that I have observed the crashes it did, in fact, crash. At that point voltage was 1.250V. The voltage then dropped to 1.000V exactly and the computer required a power cycle (same as when I had the 400W Antec so it was not the supply doing that).

So from what I understand about the overvoltage protection (OVP) it would have dropped to zero volts. So now I am thinking this is OCP (Over Current Protection) kicking in.

So now I will do some searching to see what drives the OCP and whether there is a work around. Back when I was investigating overclocking my X1950PRO I remember reading that there was a mod required to increase the switching frequency of the circuit supply Vcore so that it could supply the extra current required in OC. Maybe something similar applies here?

Another thought... Could this be a sign that I need better cooling on the MOSFETS? I just attached some heatsinks to the FETS with 3M thermal tape. But would OCP be aggravated by cooling of the FETS?

Looking for ideas. :respect:

[Solaris17]

if youd like i have a fluke meter i can measure the amps going into the card at 1.10 volts..

edit saw you had a fluke meter my bad...i just want to use it you know how expensive they are this was not a $30 multimeter http://www.tequipment.net/Fluke337A.html $350 when i bought it now $333

as for the firefly forest its heavely shader intensive thats why its loading it mroe...the leaves all move out of sink and they all have their own shadow.....as well the shadows differ with the light of firefly A and B as well as the moon C so thats 3 promary light sources that it needs to compute on top of the shimmer and shadows moving......the fet cooling is a good idea it may stabalize it more..if anything its prolonging life the OCP is aprt of the problem and their are ppl working on it iv alked to largon about it already....the theory is once we find the pin on the AL(20 more random sysmbols) chip which has 48 pins we will be able to ground the ocp pin making it non functonal the only problem is the manufactureer hasnt realesed the data sheets on the voltage regulation chip so we have no idea which one it is.

[dadi_oh]

if youd like i have a fluke meter i can measure the amps going into the card at 1.10 volts..

edit saw you had a fluke meter my bad...i just want to use it you know how expensive they are this was not a $30 multimeter

as for the firefly forest its heavely shader intensive thats why its loading it mroe...the leaves all move out of sink and they all have their own shadow.....as well the shadows differ with the light of firefly A and B as well as the moon C so thats 3 promary light sources that it needs to compute on top of the shimmer and shadows moving......the fet cooling is a good idea it may stabalize it more..if anything its prolonging life the OCP is aprt of the problem and their are ppl working on it iv alked to largon about it already....the theory is once we find the pin on the AL(20 more random sysmbols) chip which has 48 pins we will be able to ground the ocp pin making it non functonal the only problem is the manufactureer hasnt realesed the data sheets on the voltage regulation chip so we have no idea which one it is.

Yeah. Fluke is great. I bought this for ~$150 about 10 years ago. Still purring along.

Another toy that I have is a clip on current meter. I could clip it around the 12V auxillary leads to the video card but not sure how useful that is since it also gets power from the PCIX connector.

I did some more reading on another forum and this does sound like the elusive OCP grail that everyone is "currently" searching for ;)

Looking at the Vmod table I might be able to remove VID2 and drop the voltage by another 50mV to 1.150V. That might get me just under the wire and still give me a little more headroom on core frequency. Hmmmm... Time to warm up the soldering iron....

[Skitzo]

No, these are zero ohm resistors to just provide a logical low level to a digital input. Basically they feed a binary word into the input of the voltage regulating circuit. So the presence or absence of a resistor is just a binary "0" or "1" to the voltage circuit rather than an analog voltage. That is the table that i posted in one of my earlier posts.

thanks, I had two pictures in my head. That was pretty close to one of them.:)

[Solaris17]

go for it im at 1.10 did have a card at 1.15 for awhile but i returned it it ws faulty let me know how it goes. i think ill boot oblivion again im running SLI and i have oblivion running at HDR lighting and forces AA at 32Q(SLI) and AF @ 16x i think ill emasure the voltage.

[dadi_oh]

Yeah. Fluke is great. I bought this for ~$150 about 10 years ago. Still purring along.

Another toy that I have is a clip on current meter. I could clip it around the 12V auxillary leads to the video card but not sure how useful that is since it also gets power from the PCIX connector.

I did some more reading on another forum and this does sound like the elusive OCP grail that everyone is "currently" searching for ;)

Looking at the Vmod table I might be able to remove VID2 and drop the voltage by another 50mV to 1.150V. That might get me just under the wire and still give me a little more headroom on core frequency. Hmmmm... Time to warm up the soldering iron....

Set the core to 1.15V. Measured 1.160V @ idle with core at default 700MHz. Other load readings at 700MHz... ATI Tool = 1.193. Proxycon = 1.189V. Firefly Forest = 1.192V.

So I went for the gusto and set the core to 840MHz. No go. OCP kicked in as soon as turned on ATI Tool fuzzball. Next tried 820MHz and same thing. OCP.

Got it to work at 800MHz and it seems stable running 3dMark06 and artifact finder in ATI Tool. I would want to run overnight to see if it is truly stable. But I was achieving 790MHz at default 1.1V so I haven't really gained much other than some soldering practice :laugh:

So currently at:

Core: 800MHz
Shader: 2000MHz
Memory: 2140MHz

That's about it for this card until someone cracks the OCP puzzle. At least I am running cool. Only hitting around 33-34C under load.

Time to move onto something else. I should be happy with this I suppose but I think that the core has so much more potential. Seems like a poor power design is holding it back.


:toast:

[Solaris17]

congrats dude A+

[Skitzo]

nice work ... can't wait to see what happens when ocp is defeated. I'd have to say your experience has made me reconsider a solder v-mod. My speeds are pretty good at the voltages I have stock so ... yeah right, I really don't think I'm gonna be able to pass on this v-mod lol. :roll:

[dadi_oh]

Just as an epilogue to all of this....

I have come to the conclusion that the issues that I have gone through in volt-modding and overclocking this card are all related to the poor power design on the reference 9600GT boards. Those that have non-reference designs (e.g. Palit) seem to be able to volt-mod and overclock (>900MHz I have seen) without running into the extremely sensitive OCP that plagues the reference design. I assume that the non-reference designers figured out the poor reference power design and beefed it up in their designs.

With my MSI board (reference design) I have now returned to default Vcore of 1.1V. Anything over 1.1V and I will eventually get black screen (OCP) when playing games even at default 700MHz core. (There is a long string on the eVGA forum of people DOWNclocking their cards to get them to run). At 1.15V I can benchmark successfully at 815 core in 3DMark06 but if I try playing a game longer than 30 minutes or so I WILL blackscreen (OCP). If I back off to 1.1V I can also benchmark at 815 on the core and can go a little longer without OCP kicking in but I can not successfully run overnight with looping 3DMark06 without OCP eventually killing the card.

So my current status is running at 1.1V default with core at 750MHz and Memory at 2100MHz. This will run stably overnight and I can play several hours of COD4 with no OCP issues. Just a shame since with my watercooling the core is only at about 33C so there is a lot more potential left in the GPU itself. Just a poor power design on the reference 9600GT holding back some awesome performance.

Hope that helps others struggling with reference designs. My advice...

1) Don't bother with overvoltage. If it does help get higher core clocks you will never see it since OCP will kill you long before the core runs out of steam.
2) If you successfully bench at a given overclock don't assume you are OK. You should try some overnight runs with benches looping to check for stability. Or just try gaming for about an hour or two. More fun anyways :)
3) If you have a choice between a reference design 9600GT or a custom design (like Palit) then go for the custom design, particularly if you plan on overclocking and absolutely if you plan on volt-modding.

[Skitzo]

K, i can't say I am convinced the reference model wont take a little more power. My xfx idles at 1.17 volts from the factory... if I can manage to take a decent pic of it, perhaps we could compare it to another to look for differences. There must be a reason... I haven't seen mention of any other cards running at this voltage from the factory. May be worth looking into.:)

[dadi_oh]

K, i can't say I am convinced the reference model wont take a little more power. My xfx idles at 1.17 volts from the factory... if I can manage to take a decent pic of it, perhaps we could compare it to another to look for differences. There must be a reason... I haven't seen mention of any other cards running at this voltage from the factory. May be worth looking into.:)

I measured the voltages at idle and under load but don't have them handy but I think I was typically seeing about 10mV above spec at idle and about 40mV above spec under 3D Load. Your card seems to be idling about 70mV above spec unless XFX factory overvolts it. Have you measured your voltage under load? If it is hitting about 1.2V under load that would be about the same numbers I saw with my card voltmodded to 1.15V.

If XFX factory set it to 1.15V then you would see VID5 shorted, and VID2, VID3, VID4 all open. See below. Otherwise you just happen to have a card that is higher than normal. Could be just variances in component tolerances.

One of the things that is obviously different with my card is that I have added watercooling to the GPU and removed the stock heatsink. As a result I have had to add aftermarket heatsinks on the power section. Referring to the attached drawing, I added Zalman VGA memory heatsinks (the blue ones) to the 3 areas circled in red. Later on, I was examining the stock heatsink to see if I missed anything and I noticed that there was an opening in the stock heatsink that causes air to blow directly over the device circled in yellow. I put my finger on that device and it was quite a bit hotter than surround devices so I popped another heatsink on that device. I also mounted an 80mm fan on the bottom of my case pointed directly at the memory and power heatsinks so I have really good airflow over them.

It is "possible" that my card is more sensitive to OCP because my cooling of the power section is not as good as stock but I doubt it.

Just as an update, I decided last night to get a bit more aggressive on my OC so I set it to 780MHz on the core, 2020MHz on the shaders, and 2140MHz on the memory. That passed ATI Tool artifact tester so I ran it overnight with 3DMark06 looping. 8 hours later it was still running so I consider this stable. One last test would be a couple of hours of COD4. I can benchmark at 815MHz on teh core but it is not 24/7 stable.

So I can't say for certain that all cards are as sensitive as mine to overvoltage but based on the number of complaints on the web about OCP I suspect I am not that unusual. Still, I have to be pretty happy about 780/2020/2140 being 100% stable on the card.

My next project is an E8200 that is on it's way. I ordered a Swiftech MCW-30 chipset water block to put on my Northbridge since I am shooting for 500MHz FSB :eek: to get a 4GHz OC. Fingers crossed....

[Skitzo]

Yes, vid 5 is shorted. Factory voltage setup for 1.15 measured at 1.17. hitting 1.21 under load.
This is what I am referring to. from all that I have read, my card is the only that I have noticed withh this stock voltage. So I am curious what if any other changes have been made to accomodate this voltage increase. Reference designs would appear to be having minimal luck exceeding factory voltages. My card could indicate why... My factory clocks are 740/1850/2000. 740 mhz may have required a little extra voltage for stability; 740 also seems to be a bit of a ceiling for oc in a lot of cases as well. I suppose I should ask what factroy voltages refence models run at...:)

[dadi_oh]

Yes, vid 5 is shorted. Factory voltage setup for 1.15 measured at 1.17. hitting 1.21 under load.
This is what I am referring to. from all that I have read, my card is the only that I have noticed withh this stock voltage. So I am curious what if any other changes have been made to accomodate this voltage increase. Reference designs would appear to be having minimal luck exceeding factory voltages. My card could indicate why... My factory clocks are 740/1850/2000. 740 mhz may have required a little extra voltage for stability; 740 also seems to be a bit of a ceiling for oc in a lot of cases as well. I suppose I should ask what factroy voltages refence models run at...:)


Aaahh. I see. So XFX is factory over-volting the core. Interesting. So maybe they have a way to beef up their power circuit to make it less sensitive to OCP. Do they use the stock cooler design as well?

My MSI OC card is running stock 1.1V from factory and has less of a factory OC than your XFX on the core. Makes sense. And the MSI memory is factory clocked to only 1900 on memory so XFX is going a little higher there as well.

[Skitzo]

Yep, I took over 60 pics but they are no good. Between the black pcb and the low quality camera, no close ups. yes it uses the factory cooler with the full card shroud.

[Solaris17]

whanna know whats odd skitzo? when i was swapping out 9600's to get sli to work..i kept getting BFG obviously...but i got a card that had a blue PCB was stock at 1.15v and the cooler was like b4 release cooler...instead of shiny metal it was like ..matted black paint...and do you know on the top side of the cooler how it says nvidia and has the logo printed on the side in chrome? ya didnt exist..it was the oddest card i ever saw so i spoke to bfg and he said he had no idea he ahd never seen it b4.

[dadi_oh]

Yep, I took over 60 pics but they are no good. Between the black pcb and the low quality camera, no close ups. yes it uses the factory cooler with the full card shroud.

So XFX is doing the Clint Eastwood..... "You gotta ask yourself one question... Punk. Ya' feelin' lucky?" :)

Would be interesting to know if the XFX forum has statistically more OCP type issues than other manufacturers. I think it was the eVGA forum that had people complaining about needing to underclock their cards from factory settings to get them to run without black screens.

[Skitzo]

i don't like the cooler... doesn't fit well.

[Skitzo]

I tried to do the vmod with a pencil but it didn't work. I don't really want to void the warranty if the mod won't work. I need to make a temp connection for testing. Any ideas? with vid 5 shorted, i need to short vid2 to bump it up .05V (1.20V)

[dadi_oh]

I tried to do the vmod with a pencil but it didn't work. I don't really want to void the warranty if the mod won't work. I need to make a temp connection for testing. Any ideas? with vid 5 shorted, i need to short vid2 to bump it up .05V (1.20V)

Yup. VID5 and VID2 shorted. VID3 and VID4 open. = 1.2V.

There is a great guide here on this forum that shows all the options.

Let us know if your OCP creates issues at 1.2V. The best bench to cause failure according to my experiments is Firefly Forest in 3DMark06. Lot's of shader work and my voltage measurements indicate the Vcore peaks during this test. Put it into a continuous loop on that test and if the OCP is marginal it should show up within a couple of hours.

[Skitzo]

any ideas on a temp connection?
wonder if an aligator clip and a piece of wire will work. i know, don't short anything else ... lol. There has to be a way...:)

[dadi_oh]

any ideas on a temp connection?
wonder if an aligator clip and a piece of wire will work. i know, don't short anything else ... lol. There has to be a way...:)

Sorry. I'm replying to you in 2 different threads :)

Temporary hmmmm.... I have always used a soldering iron but obviously that voids warranty.

I have heard some people use the conductive ink for repairing rear windshield defrosters. They claim it comes off with a little isopropyl alcohol (on the ink that is... not down the gullet :-)

[Skitzo]

lol i'm asking in two threads so ... yeah I was looking at the repair kit(rear defroster auto) The one I found looked like copper not silver. I think it was 12.99 for 15 ml. I can get a 250 ml bottle of silver conductive paint for 30.00. Wonder if there are any major diferences aside from color and price.:)

[Skitzo]

whanna know whats odd skitzo? when i was swapping out 9600's to get sli to work..i kept getting BFG obviously...but i got a card that had a blue PCB was stock at 1.15v and the cooler was like b4 release cooler...instead of shiny metal it was like ..matted black paint...and do you know on the top side of the cooler how it says nvidia and has the logo printed on the side in chrome? ya didnt exist..it was the oddest card i ever saw so i spoke to bfg and he said he had no idea he ahd never seen it b4.

cool, still have it?:)