eVGA 9600GSO Dual-Slot volt mod?

[quadomatic]

I'm looking to volt mod my eVGA 9600GSO Dual-Slot Edition card. It's the one that went on sale for $34 ($31 if you did Live Search Cashback) back in September. The card doesn't follow the reference model, so I took the cooler off and took some pictures of the bare card.

It seems like I can overclock this card quite a bit. Without the volt mod, it runs at 752 mhz core/1806 mhz shader/1050 mhz memory. No problems as of yet. When I took the cooler off I found that the memory chips are Samsung, not the qimonda chips that were mentioned as being questionable earlier (though I don't know anything about the Samsung chips).

For $31, it definitely seems like I'm getting my money's worth. Fallout 3 runs beautifully, as does most every other game I've tried (Crysis is a bit questionable, but that's to be expected).

So, if someone could have a look at these pictures and tell me what I can do for a volt mod, I'd appreciate it:



Back:

http://img526.imageshack.us/img526/1691/backegs3.th.jpg (http://img526.imageshack.us/my.php?image=backegs3.jpg)

Zoomed (Left, Center, Right):
http://img526.imageshack.us/img526/632/backleftefw8.th.jpg (http://img526.imageshack.us/my.php?image=backleftefw8.jpg)http://img440.imageshack.us/img440/3953/backcentereph5.th.jpg (http://img440.imageshack.us/my.php?image=backcentereph5.jpg)http://img444.imageshack.us/img444/3251/backrighterb6.th.jpg (http://img444.imageshack.us/my.php?image=backrighterb6.jpg)

Front:

http://img440.imageshack.us/img440/1153/fronteuc0.th.jpg (http://img440.imageshack.us/my.php?image=fronteuc0.jpg)

Zoomed (Left, Center, Right):

http://img243.imageshack.us/img243/316/frontleftear3.th.jpg (http://img243.imageshack.us/my.php?image=frontleftear3.jpg)http://img442.imageshack.us/img442/231/frontcentereqs8.th.jpg (http://img442.imageshack.us/my.php?image=frontcentereqs8.jpg)http://img440.imageshack.us/img440/2646/frontrightelp4.th.jpg (http://img440.imageshack.us/my.php?image=frontrightelp4.jpg)

Thanks a bunch!

[sniperbob]

Kinda off topic, but thanks for posting the pictures of the card. I have two of these exact same cards and am looking into water-cooling options, so these stripped pictures are pretty valuable to me as to how to plan my liquid attack! Good luck figuring out the voltmod and sharing your clocks. My cards are set to 651 core/ 1842 shader/ 853 memory with fan speed cranked to 100% all the time. I use my rig for folding so they are under 100% load all the time and i have yet to have any problems with my clock. I may try bumping my cards up more after seeing your mem and core clock settings. :up:

[quadomatic]

There, they've been uploaded to Imageshack...I had to reduce quality of the images from 89 to 85 in GIMP though.

Back:

http://img526.imageshack.us/img526/1691/backegs3.th.jpg (http://img526.imageshack.us/my.php?image=backegs3.jpg)

Zoomed (Left, Center, Right):
http://img526.imageshack.us/img526/632/backleftefw8.th.jpg (http://img526.imageshack.us/my.php?image=backleftefw8.jpg)http://img440.imageshack.us/img440/3953/backcentereph5.th.jpg (http://img440.imageshack.us/my.php?image=backcentereph5.jpg)http://img444.imageshack.us/img444/3251/backrighterb6.th.jpg (http://img444.imageshack.us/my.php?image=backrighterb6.jpg)

Front:

http://img440.imageshack.us/img440/1153/fronteuc0.th.jpg (http://img440.imageshack.us/my.php?image=fronteuc0.jpg)

Zoomed (Left, Center, Right):

http://img243.imageshack.us/img243/316/frontleftear3.th.jpg (http://img243.imageshack.us/my.php?image=frontleftear3.jpg)http://img442.imageshack.us/img442/231/frontcentereqs8.th.jpg (http://img442.imageshack.us/my.php?image=frontcentereqs8.jpg)http://img440.imageshack.us/img440/2646/frontrightelp4.th.jpg (http://img440.imageshack.us/my.php?image=frontrightelp4.jpg)

[panzerchaos47]

I have the same cards. Trying to find a voltmod to overclock them higher than 1800 (folding) but overclock.net has the early rev. If you can post the names of the those chips (back of the card 1 in the left and 2 in the mid right of the card). It might help.

[quadomatic]

Not sure what you mean by mid right, but on the front is an RT8802A, and that regulates the core.

As for the one on the back left, that's AJ-AA UCP...I can't read it. It's so tiny.

[panzerchaos47]

Oh never mind didn't know the chip was on the front. Just wait for someone that could help.

[quadomatic]

Infrared on TechPowerUp found the data sheet for the chip. Based on what he found, he says that by removing certain surface mount resistors, the voltage can be raised by certain amounts, by which the voltage to the core can be raised.

[panzerchaos47]

Thanks!! I think I got it.
If I wanted to get 1.25v I have to cut VID 0, VID 1 and VID 2? That's what he was saying.

[quadomatic]

Not sure, but here's the thread:

http://forums.techpowerup.com/showthread.php?t=53910&page=3

Scroll towards the bottom and look at infrared's posts.

Also here's the chip's datasheet:

http://pdf1.alldatasheet.com/datasheet-pdf/view/207883/RICHTEK/RT8802A.html

I'm not sure what the default voltage of the card is. Do you know?

http://img145.imageshack.us/img145/2614/img1872emd1.th.jpg (http://img145.imageshack.us/my.php?image=img1872emd1.jpg)

I'm looking at that picture to determine default voltage and which resistor connections go to which pins.

It LOOKS like VID1 and VID6 are the only pins that don't have resistors, so those pins aren't connected to anything (meaning 0 I guess?), so I guess the default voltage is 1.11875V.

So, it looks like in order to raise voltage, do you not remove resistors and bridge gaps, but just make gaps?

UPDATE: I think this is the proper layout:

http://img519.imageshack.us/img519/4944/chipql1.jpg (http://imageshack.us)
http://img519.imageshack.us/img519/chipql1.jpg/1/w642.png (http://g.imageshack.us/img519/chipql1.jpg/1/)

[panzerchaos47]

Looks like when I was trying to mod the bios the highest voltage is 1.1v
I might try it out but there is no actual measure points unless you know the accurate one.
---
Updated on Nov 17 13:29
If you want the pencil mod
If you look to the upper left chip and 4th pin (Find the FB pin on datasheet. You trace to the resister and measure it you should get 3.92 kohm. Shade that resister about 3.70 kohm. The temperature will increase by 6 - 7 C

GPU0 without pencil mod is 56 C max on hwmonitor
GPU1 with the mod is 62 C

I tested with folding which is inaccurate (Looks like easier to test the shaders for me if there is a good test)
GPU0 is clock at 680 / 1800 / 1000
GPU1 is clock at 680 / 1860 / 900 stable for 22 mins so far

I used this guide http://www.aoaforums.com/forum/hardware-hacking/39385-diy-volt-mod-guide.html
Ill try to find that vcore measurements.

[quadomatic]

Looks like when I was trying to mod the bios the highest voltage is 1.1v
I might try it out but there is no actual measure points unless you know the accurate one.
---
Updated on Nov 17 13:29
If you want the pencil mod
If you look to the upper left chip and 4th pin (Find the FB pin on datasheet. You trace to the resister and measure it you should get 3.92 kohm. Shade that resister about 3.70 kohm. The temperature will increase by 6 - 7 C

GPU0 without pencil mod is 56 C max on hwmonitor
GPU1 with the mod is 62 C

I tested with folding which is inaccurate (Looks like easier to test the shaders for me if there is a good test)
GPU0 is clock at 680 / 1800 / 1000
GPU1 is clock at 680 / 1860 / 900 stable for 22 mins so far

I used this guide http://www.aoaforums.com/forum/hardware-hacking/39385-diy-volt-mod-guide.html
Ill try to find that vcore measurements.

I found the FB pin. We are talking about the same chip as before right? The RT8802A?

By shading the resistor, that means shading all over the resistor so that the current passes over the resistor rather than through it right? You don't have to trace to the resistor with pencil do you? Just shade the resistor right?

Could you send me a picture of what you did?

Instead of using folding, try using ATITool artifact scanner. Use eVGA Precision Tools to overclock, and then run the artifact scanner in ATITool.

Are you sure you can't push core clocks more than that? Mine ran at 752 before even doing a volt mod. And so was memory not stable and you had to lower mem speeds? Maybe because whatever thermal material that was on there was removed when taking off the cooler?

Hmmm...instead of doing it through the FB pin, couldn't we just do the same thing through VID pins? Like, shade the VID3 resistor?

Then again I'm not sure if that's such a good idea. According to what you said now, based on the change in resistance, the voltage was probably 1.165V after you did that. Raising it to 1.32V might totally screw the card...but is that formula given in your link applicable to the pencil mod? Maybe the voltage is higher? Would GPU-Z tell you?

Does the increase in voltage depend on how much you shade the resistor?

Or wait...I don't think that mod would work. The idea isn't to shade over the resistor with the VID resistors but to remove them and make a gap in the connection. Right?

[panzerchaos47]

Red is the resistor that i shade the pencil
Blue is the measure the ohm starts 3.92 Kohm

I meant to trace it to find the resistor from FB pin since that time I didn't see option to save the picture until now. lol

I used your picture since it's the same and I don't have a camera that take a good picture.
http://i256.photobucket.com/albums/hh166/hitokirixii/9600gsopencilmod.jpg
Photobucket reduce the size of the picture.

Only problem is finding the vcore measure point. I think I'm close to finding it.

Here ones don't know if this is accurate though. Hope this helps.
For GPU0 I got 1.25 on idle and around 1.28v to 1.30v on load
For GPU1 I got 1.32 on idle and around 1.34v on load
http://i256.photobucket.com/albums/hh166/hitokirixii/9600gsomeasurepoint.jpg

[quadomatic]

Red is the resistor that i shade the pencil
Blue is the measure the ohm starts 3.92 Kohm

I meant to trace it to find the resistor from FB pin since that time I didn't see option to save the picture until now. lol

I used your picture since it's the same and I don't have a camera that take a good picture.
http://i256.photobucket.com/albums/hh166/hitokirixii/9600gsopencilmod.jpg
Photobucket reduce the size of the picture.

Only problem is finding the vcore measure point. I think I'm close to finding it.

Here ones don't know if this is accurate though. Hope this helps.
For GPU0 I got 1.25 on idle and around 1.28v to 1.30v on load
For GPU1 I got 1.32 on idle and around 1.34v on load
http://i256.photobucket.com/albums/hh166/hitokirixii/9600gsomeasurepoint.jpg

I see.

What exactly is GPU0 and GPU1?

[panzerchaos47]

I have two 9600 GS0 so I label both GPU0 without pencil mod and GPU1 with pencil mod. I wish I knew better measure point because I thought it was 1.118v was the default what you pointed out.

[quadomatic]

I don't think that's likely the actual point though, since it registers 1.25V on idle w/out pencil mod, and BIOS limits voltage to 1.1V by default, and based on the VID diagram I figured the voltage is 1.11875V.

Either way, it seems to accurately predict the change in voltage.

Based on calculations of resistance changes measured with the multimeter, the voltage change should be about +.07V...and 1.32V-1.25V=.07 So....I guess that point is at least reliable for measuring CHANGE in voltage.

Does the change in voltage using the FB pin depend on the thickness of the shading over the resistor?

Oh and did you try raising core clock speed more? You should be able to push it a bit more than that.

[quadomatic]

I decided to try the FB volt mod. It doesn't seem to be producing the desired effect. I had to lower clock speeds back on the memory. Maybe I shaded to thick? It seems like the gpu is hotter than it used to be...is 64 odd?

Maybe because I didn't reapply thermal grease properly? I just shaved off the dried paste that was on the heatsink before and threw it back onto the core. Probably has some burn in time, but still...

Maybe the added voltage requires a better cooler because of the extra heat?

[panzerchaos47]

I don't know I tested it again for some reason I think the my GPU1 did good with overclocks at 3.7 kohm so I did pencil mod for 3.4 kohm on GPU0. It got hot but not very hot around 60 C for GPU0 used to be around 54C. Couldn't reach more than 1800 Mhz on shaders while the GPU1 could reach to 1900 without messing up.
Let me continue to work with the GPU0. I'll focus on one 9600 GSO.

[quadomatic]

I think the reason why my card got so hot was because the thermal paste was no good. I took the cooler off of the card to take a look and it hadn't gone from the dry putty state it had been to a smoother state, so I'm assuming that the stuff had oxidized after I took the card out, and it was foolish to expect that I didn't need to put new paste on...

Right now I'm looking to buy an Accelero S1 and to strap a 120mm fan to it with a zip tie. I'm not sure whether old Accelero S1 coolers work on 8800GT though.

UPDATE: I decided to go for it. I picked up an S1 on eBay for $17 after Live Cashback. This cooler is supposed to be pretty amazing (I already have Accelero S2, but unfortunately it doesn't fit on this card), so I'll strap a 120mm fan onto it and hopefully this will help.

[panzerchaos47]

I did a lot more test on GPU0 with and without the pencil mod. The I shaded the resister to 3.16 kohm there was no difference. I guess it was because reapplying thermal grease so many times sometimes temperature changes (somtimes 60 or 70). I still can't get the overclock with over 1800. I think my 2nd GSO was with luck. Also found out the measure points isn't working. Doesn't give a constant reading like 1.35v then later reboot it could do 1.5v so don't use that. Getting the better vcore measure would help a lot. I wish it was easy as the HD 3870.

[quadomatic]

Perhaps this mod doesn't actually work then? Maybe the only way is through the VID pins?

But, then again, I was able to run at higher clock speeds and memory speeds without it immediately crashing on me. It was only a little afterwards that I got artifacting. I think one of the reasons was because the memory wasn't being adequately cooled (notice those styrofoam pads?). They had some liquid thing on them before I took off the cooler. That wasn't there anymore when I went to put the cooler back on. I'm guessing whatever that was, was keeping them at least a little cool or something. Maybe it helped to transfer heat.

[panzerchaos47]

Yeah I guess it won't work since other GSO vmods didn't mention any pencil mods but for the memory.

[quadomatic]

We'll see for sure once I replace the cooler. If I find that I'm able to break 760 on core without it crashing (it used to do this immediately, not because of heat) then I'd be pretty sure it works.

[panzerchaos47]

I found out there was vmod for vmem which is kind of useful since this card has limited memory bandwidth.

Same as what I did but it's for APW7067N that 14pin on top of the penny. Look at datasheet for FB pin then shade resister. If you measure it stock is 591 (or near) ohms.
http://pdf1.alldatasheet.net/datasheet-pdf/view/149550/ANPEC/APW7067N.html

Voltage measurement is very accurate. Check same picture that I posted it's lower right of the green arrow.

590 ohms = 1.88v / 1.90v
572 ohms = 1.93v / 1.94v
556 ohms = 1.97v / 1.98v

This first successful vmod I did but for vmem.

[quadomatic]

Awesome! You're right about that. Increased memory speeds would probably make a lot more of a difference. Plus these are samsung chips, not the nasty qimonda, so we might be better off.

DOM (Palit 9600GSO http://www.teampalit.com/showthread.php?t=6661) got to 1150 with his memory. I got to 1050 without a volt mod, so this could help quite a bit.

How much more were you able to raise the speeds? I'd suggest you get an Accelero S1 since you would then be able to put heatsinks on the memory so they'd actually get cooled some.

Does the new voltage depend on how thickly the resistor is shaded?

Now if only we could figure out the core...not sure how much we'd be able to push it though. If it can get to 850 I'll be pretty dang impressed, though that seems really unlikely on air. On water, that's a different story...

I wish largon would respond to my PM's...he'd probably be able to help quite a bit.

UPDATE: Is this the resistor that needs to be shaded for vmod on mem?

http://img136.imageshack.us/img136/3610/vmemqq4.jpg (http://imageshack.us)
http://img136.imageshack.us/img136/vmemqq4.jpg/1/w534.png (http://g.imageshack.us/img136/vmemqq4.jpg/1/)

[panzerchaos47]

That's wrong side though. Sounds the same thing like that other 9600 GSO mods from techpowerup that has no pencil mod for vcore but different PCB.

Vmod for Vmem. I was able to get it over 1050 without crashing. Measure it before you do shade it. Should be around 590 ohms.
http://i256.photobucket.com/albums/hh166/hitokirixii/9600gsovmem.jpg

Blue is measure point for Vmem
http://i256.photobucket.com/albums/hh166/hitokirixii/9600gsomeasurepoint.jpg

[quadomatic]

I got some OCZ Freeze and put the old cooler back on the card. The temps dropped a good 10 degrees on load from before. The volt mods still don't appear to be doing it for me.

I shaded the memory resistor but i haven't gotten anything out of it. In fact I had to lower mem speeds to 1037. It's probably because this cooler doesn't actually provide adequate cooling to memory.

We'll see how it works out once I get the Accelero S1 on Tuesday and can actually put heatsinks on the memory, as opposed to the stupid styrofoam pads that come with the card. It seems like whatever ability the cooler HAD to cool the memory isn't there anymore (there was some liquid sitting on the memory chips when I took the cooler off, but that wasn't there by the time I put the cooler back on, which was at least a couple weeks later.)

[CL3P20]

It seems, on the data sheet for the RT8802A, that FB is independent of both the DAC and droop circuits...so shading FB resistor, or FB to VR should not increase GPUv at all. I have been pouring over this and various threads for a while now..[i have 8 GSO's in front of me awaiting mods]. It looks to me that 'Adj' on pin 15 should tune vdroop.. I might test this out after a confirmed vcore read..

Any luck yet, I will be removing VID2 for 1.218v idle..I am gonna have to put one of these on the MCW60 to be able to prod around easier..

[quadomatic]

Awesome, good to have another person snooping around on this thing. I'm still not sure that the memory volt mod was actually successful, since I wasn't able to push memory any more than I had (and had to even drop the speeds some).

How would you find a vCore read point? Also, the card has to be on to be read right?

[CL3P20]

Edit:

GPUv I am using is on the back of the large inductors :thumb:

[panzerchaos47]

Wow thanks I was doing electrical tape and floppy wire to test those mosef but not the two legs. I did a lot of reapplying of thermal grease and putting heatsink. I might do this if I have time.

----
What was ohms when you did the pencil mod on vmem?

--Uodate 11/26--
I don't know if this Samsung memory is really bad at overclocking I need like 2.17v for 1100 Mhz. Anyone know what's good safe voltage for this memory?

I tried the measure points so far I get .9 on middle leg of the mosef while left leg is 9.9 (Don't remember if it's the top or bottom mosef)

[CL3P20]

I tried removing VID resistors #3 and 4..for 1.21v on one of my cards.. strange..the card works fine..but as far as I can tell..the mod has no effect. Temps, and clocks are the same as before the mod. Only vmeasure I can find reads @ 1.04v..which makes sense if default vcore for my cards is ~1.08v..that being load.. ~.04v less for idle.
I am just gonna have to try something else... I think pin40 'VIDsel' has something to do with it.. maybe VR will cause something to happen. At this point I'd be satisfied with change..good or bad. :o

[quadomatic]

I tried removing VID resistors #3 and 4..for 1.21v on one of my cards.. strange..the card works fine..but as far as I can tell..the mod has no effect. Temps, and clocks are the same as before the mod. Only vmeasure I can find reads @ 1.04v..which makes sense if default vcore for my cards is ~1.08v..that being load.. ~.04v less for idle.
I am just gonna have to try something else... I think pin40 'VIDsel' has something to do with it.. maybe VR will cause something to happen. At this point I'd be satisfied with change..good or bad. :o

Hmph, yeah...this is really odd. Nothing we've tried has worked so far.

Maybe we need to get someone else to look at it as well. It'd certainly help...

[quadomatic]

So I put on my Accelero S1, finally. I strapped a 120mm fan onto it as well. It's WORLDS quieter than the 9600GSO dual slot fan at 100%. Temps have dropped a sizable amount, and I was able to raise clocks up again.

Now, my card runs at 741/1836/1053. That's quite a bit higher than I was able to run the shader clock before (clocks before were 741/1782/1037), so I'm happy with the increase. Also, temps seem to peak at 53 or so in artifact scanner. I can't push anything further than this though, or else I get some decent artifacting.

Any clue whether we could just try a bios vmod for kicks to see?

[CL3P20]

BIOS might actually work..especially since I cant seem to make heads-or-tails of the rest of the datasheet. :( According to the datasheet, the IC is compatible with 3 different types of board layouts..the nomenclature for BIOS GPUv is listed in the data sheet..If you are handy with BIOS flashing, you might give it a try.

*I will solder the crap out of something before I attempt a BIOS flash..I am a BIOS flash noob. :D

[quadomatic]

BIOS might actually work..especially since I cant seem to make heads-or-tails of the rest of the datasheet. :( According to the datasheet, the IC is compatible with 3 different types of board layouts..the nomenclature for BIOS GPUv is listed in the data sheet..If you are handy with BIOS flashing, you might give it a try.

*I will solder the crap out of something before I attempt a BIOS flash..I am a BIOS flash noob. :D

People have said that 8800GS and 9600GSO have voltages limited to 1.15V though. When I looked at the datasheet for VID resistors, it looks like the card was limited to 1.11875V though. I can take a look though at bios, though I'm still a flash n00b.

We're really going to need help if we're going to go the hard mod route though.

[quadomatic]

I tried removing VID resistors #3 and 4..for 1.21v on one of my cards.. strange..the card works fine..but as far as I can tell..the mod has no effect. Temps, and clocks are the same as before the mod. Only vmeasure I can find reads @ 1.04v..which makes sense if default vcore for my cards is ~1.08v..that being load.. ~.04v less for idle.
I am just gonna have to try something else... I think pin40 'VIDsel' has something to do with it.. maybe VR will cause something to happen. At this point I'd be satisfied with change..good or bad. :o

WAIT!!!!!!

There is no mod that has VID3, VID4, VID1, and VID6 removed that would put it at 1.21v. Removing VID3 and VID4 would put it at 0.94375V, since VID1 and VID6 aren't in by default. So maybe we aren't moving in the wrong direction with removing resistors.

I'm pretty sure this is a discrepancy in how we read the data sheet.

I think 0's mean the points are not connected, and 1's mean points are connected. If we read it the way you did, and say 0's mean points are connected, and 1's are not connected, then default voltage is 1.03750v. I don't think that's right, since when panzerchaos looked at bios he said the highest voltage was 1.1v, which is what I read it as with 0's not being connected 1's being connected.

So instead, if you think that the way I read is right instead, and if you put back VID3 and VID4, removed VID2 and see what voltage that gives you.

Do we need to maybe mod the bios after removing the resistors as well?

[CL3P20]

I think 0's mean the points are not connected, and 1's mean points are connected. If we read it the way you did, and say 0's mean points are connected, and 1's are not connected, then default voltage is 1.03750v. I don't think that's right, since when panzerchaos looked at bios he said the highest voltage was 1.1v, which is what I read it as with 0's not being connected 1's being connected.

the vcore read I am using indicates 1.04v at idle..before and after the resistors were removed..clocks and temps reflected the same...

..pretty sure it says on the datasheet which digit indicates 'open' or 'grounded'.. cant open PDF right now, running on stripped bench OS ATM.

[quadomatic]

the vcore read I am using indicates 1.04v at idle..before and after the resistors were removed..clocks and temps reflected the same...

..pretty sure it says on the datasheet which digit indicates 'open' or 'grounded'.. cant open PDF right now, running on stripped bench OS ATM.

I think you're right, since I found it in the datasheet now, but specifically for that chart. Still doesn't explain what panzerchaos found with the 1.1v in bios.

1.04v is pretty low...I think others were able to do 1.15v by bios mod on theirs. If the chip sets it to 1.04v, does that mean its hard locked to 1.04v? And we can't raise it more than that by bios?

Maybe there's some regulator that needs to be modded too, along with the resistors removed.

[quadomatic]

I just finished doing a bios mod. I backed up the bios from the card using nvflash, and 1.1V was listed, so I'm not sure how that fits in with the 1.04v. I changed voltage to 1.1V in nibitor and flashed the card. Doing the mod did nothing for me. I wasn't able to raise any speeds without seeing artifacts, just as before.

Can artifacts be volt related? Either the artifacts I get right now when I push clocks more are related to a lack of voltage, or this card just has a wall. They aren't temperature related, since at 48 degrees I get artifacts if I push clocks further than what they are now (741/1836/1053).

[quadomatic]

http://forums.techpowerup.com/showthread.php?p=1095130

I made a thread on TechPowerUp that pretty much compiles all of our findings into one post. Hopefully it attracts some help.

I'm going to try shading the memory resistor again and try to do it right this time. Maybe I used a bad pencil.

[CL3P20]

i have been going over this datasheet with an electronics 'buddy' ...he said it seems like the output and input current portions are separately controlled..

ie- the mods are for the input side of the circuit..like 'vsense'..but no matter the input resistance..the output is locked to a certain limit by another IC or diode..

In his opinion..two mods will most likely be needed to regulate voltage manually...still working on it though. :(

[quadomatic]

Any updates?

[CL3P20]

None for me..future is looking dismal.

*Largon suggested to mod VR to pin #4 of the RT88 chip for GPUv.. but I am pretty sure this has already been confirmed as a 'no-go' for GPUv. I have tried a few other things..shading resistors linked to vdroop and OCP..but nothing. :(

[quadomatic]

None for me..future is looking dismal.

*Largon suggested to mod VR to pin #4 of the RT88 chip for GPUv.. but I am pretty sure this has already been confirmed as a 'no-go' for GPUv. I have tried a few other things..shading resistors linked to vdroop and OCP..but nothing. :(

Dang....is this just a coincidence or do they just HATE volt modders lol...

Any ideas on a memory vmod? That's where this card is lacking most, so it'd probably be the most helpful. I tried the pencil but it didn't appear to do anything for me.

[CL3P20]

From panzer's post on pg#1..


I found out there was vmod for vmem which is kind of useful since this card has limited memory bandwidth.

Same as what I did but it's for APW7067N that 14pin on top of the penny. Look at datasheet for FB pin then shade resister. If you measure it stock is 591 (or near) ohms.
http://pdf1.alldatasheet.net/datashe.../APW7067N.html

Voltage measurement is very accurate. Check same picture that I posted it's lower right of the green arrow.

590 ohms = 1.88v / 1.90v
572 ohms = 1.93v / 1.94v
556 ohms = 1.97v / 1.98v

This first successful vmod I did but for vmem.

..does this mod not work..I hadnt bothered with it yet [been busy trying to work GPUv], but was planning too..

[quadomatic]

From panzer's post on pg#1..



..does this mod not work..I hadnt bothered with it yet [been busy trying to work GPUv], but was planning too..

It doesn't appear to work for me.

[largon]

This thread is doomed and pencils are to blame. Mods work for some, but not for some. Makes no sense.
People trying to vmod without using a DMM or even without actual measuring points?
:banana:


Anyways, here are the measuring points:

92122

Someone should measure idle/load voltages from these points...

vMEM mod:

92124
Pin#7 is ground, eg.

vGPU mod:

92125

vMEM definitely works, vGPU may cause OVP tripping. Testing needed.
Alternative vGPU mod soldering point would be the pad immediately below the yellow dot.

Here's a vGPU mod that sets VID to 1.31250v:

VID mod pulled for now.


edit:
There is no reason why FB method wouldn't work. RT8802 has been modded succesfully in the past via the FB pin. The "issues" brought up in post #27 are present in all cards.

edit2:
VID mod pulled.

[quadomatic]

Largon to the rescue! Thanks a ton.

I'll get a hold of some variable resistors and give the vMem mod a shot first, since memory is the bigger issue with this card.

Do you know where the ground is for the vGPU mod? The datasheet puts GND on pin 41, but it's labeled on top of the chip...kind of odd.

http://www.alldatasheet.com/datasheet-pdf/pdf/207883/RICHTEK/RT8802A.html

Figure I'll order these then:

10kohm for vMEM: http://www.mouser.com/Search/ProductDetail.aspx?qs=sKxHajcpPaC%252bknkV8v6zaQ%3 d%3d
100kohm for vGPU: http://www.mouser.com/Search/ProductDetail.aspx?qs=sKxHajcpPaD6mD8hAI4JaQ%3d%3d

I ended up ordering from Newark instead...lower prices, lower shipping cost and the VRs aren't oos there.

[largon]

Any ground will do.

[quadomatic]

Any ground will do.

So can I just solder them both to pin 7 on the chip for vmem mod?

[CryptiK]

Yes PIN 7 is a ground of the APW7067N chip. Any ground will do fine, however if you are not so good at soldering, choose a larger more easily accessible ground like one of the grounds of the 6 pin PCI-E power connector.

You should also buy a DMM, even a cheap one, there is no way you should do the mods without it. You need to find alternative grounds, test your soldering work before you power on the card, and use it to measure voltages after you have done the mods.

[quadomatic]

Yes PIN 7 is a ground of the APW7067N chip. Any ground will do fine, however if you are not so good at soldering, choose a larger more easily accessible ground like one of the grounds of the 6 pin PCI-E power connector.

You should also buy a DMM, even a cheap one, there is no way you should do the mods without it. You need to find alternative grounds, test your soldering work before you power on the card, and use it to measure voltages after you have done the mods.

Where can I pick up a cheap one?

Is this good enough?

http://www.amazon.com/Palm-Size-Handheld-Digital-Multimeter-DT830B/dp/B00066ZZO4/ref=sr_1_4?ie=UTF8&s=hi&qid=1231126761&sr=1-4

[CryptiK]

Depends where you live, but places like radioshack stock them, have a search online or in electronics shops where you live. A cheap one can cost around $10 - $20 and will be fine for what you want to do.

[quadomatic]

I went ahead and ordered that one from Amazon last night. My VRs should arrive in the next couple days too.

[CL3P20]

I just finished GPUv mod as posted.. somethings definitely not right..cant tell if its OCP or not. Rig wont even POST with GPU, so I have no voltage reading at all.. I re-soldered to a different ground location..still same results.

-is setting new VID the only way to mod for OCP?

[quadomatic]

I just finished GPUv mod as posted.. somethings definitely not right..cant tell if its OCP or not. Rig wont even POST with GPU, so I have no voltage reading at all.. I re-soldered to a different ground location..still same results.

-is setting new VID the only way to mod for OCP?

Darn...I hope vMem mod still works.

[CryptiK]

I just finished GPUv mod as posted.. somethings definitely not right..cant tell if its OCP or not. Rig wont even POST with GPU, so I have no voltage reading at all.. I re-soldered to a different ground location..still same results.

-is setting new VID the only way to mod for OCP?

Did you set the VR to max resistance before soldering it on? Are your solder joint touching anything they shouldn't be?

Yes theoretically if you remove the 2 resistors largon indicated for the VID mod it should also move the OCP/OVP (I'd guess it's OVP as there is very little current being drawn at idle) trip point up accordingly.

If you do decide to try and remove them, be careful and gentle doing so. They are quite easy to break.

[quadomatic]

Did you set the VR to max resistance before soldering it on? Are your solder joint touching anything they shouldn't be?

Yes theoretically if you remove the 2 resistors largon indicated for the VID mod it should also move the OCP/OVP (I'd guess it's OVP as there is very little current being drawn at idle) trip point up accordingly.

If you do decide to try and remove them, be careful and gentle doing so. They are quite easy to break.

Does it have to be set to max resistance before soldering? Can't you change it afterwards?

Also, why does removing resistors fix the OCP issue, but not changing voltage through the FB pin?

[CryptiK]

Yes it should be set to max resistance before soldering on, as this puts the vGPU close to standard, and then you decrease the resistance of the VR to increase the vGPU. If you have the VR resistance set too low before soldering on, it may be trying to set the vGPU above the OVP trip point and not allowing the card to turn on.

Removing the resistors and setting the VID to ~1.3v as I said *should* increase the OVP trip point, but the OVP trip point may also be set by a separate R/C circuit and the VID resistor removal may put the vGPU above this. I have yet to really have a good look at the circuit (I should get time to do this tonight sometime), largon may be familiar enough with it to suggest something here.

[largon]

VID determines OVP trip point which according to the datasheet is typically VID+170mV so setting VID->1.31250v will never trigger OVP. FB mod can and will eventually trigger OVP.

OCP will never give problems in idle. OCP triggers only during load.

CL3P20,
Please measure resistance between your soldering points with the mod installed.
And also, resistance between yellow dot (fb pin) and any of the pads left of the yellow dot:

http://www.xtremesystems.org/forums/attachment.php?attachmentid=92125&d=1231065349

[CL3P20]

resistance with mod- is reading the same Ohms as the VR not hooked up :( -> .105k Ohms I have tried 3x different ground locations now..all with same results.

Measuring between the above 'yellow' and resistor directly below it [in pic] = .105k
Measuring between the above 'yellow' and resistor directly above it [in pic] = .0000k

*I made sure to preset max resistance on the VR, before installing too .... btw, stock resistance between yellow and ground = 3.94k ohms

[CL3P20]

pic with first ground position I tried..
http://i265.photobucket.com/albums/ii220/CL3P20/P1050543.jpg

..all connections double checked.. ohms still at .105k as soon as I connect VR.. am I retarded or what..? This should work right..?

[quadomatic]

..all connections double checked.. ohms still at .105k as soon as I connect VR.. am I retarded or what..? This should work right..?

Is it possible that the VR is broken? Even if this mod wasn't workable because of OVP tripping, you should still be able to set a higher resistance, right?

[largon]

You sure that resistor with the yellow dot isn't shorted?
edit:
Does that VID (which is it, 1.21250V or 1.26250V?) work without the FB mod?

[CL3P20]

@ quadromatic- I just bought 8x of the VR's yesterday..took it out of the package..straight to the PCB.. I suppose it could be possible though.

@ largon- the resistor measures 3.9k as soon as I remove the ground from the VR.. which is original resistance value, so- Yes I am sure the resistor is not bridged or shorted. I also tried pin7 on the 7067N for another ground..same results..just wanted to make sure. I will swap VR's out with a new one..and report back..

[largon]

Wait, what?
Are you saying that the same spots that measure 3.9kΩ without the mod measures 0.105k (~105Ω) when the mod is in? That is, resistance between FB and ground is 3.9kΩ without the VR and 0.105k with it?

Are you sure you're using a 100kΩ (100 000Ω) VR as instructed? Sounds like it's 100Ω.

[CL3P20]

OMG.. you are so right.. I have waaayy to many VR's laying around here.. must have grabbed the 100 by accident *doh*.. god I feel like an idiot right now..
Thanks for the correction, Largon.. I appreciate your patience, with my ignorance..

TY for helping a fellow 'Fin'. :D

[largon]

Hehe. Well, we'll see if I've been of any use after you get the correct VR in there - as in does the mod work. ;)
Am I right you've already done some vGPU VID mods? How do they work?


TY for helping a fellow 'Fin'.You got finnish roots?

[quadomatic]

Awesome. Sucks that the mistake was made but it still leaves hope. Let us know how it works out.

[CL3P20]

Hehe. Am I right you've already done some vGPU VID mods? How do they work?

You got finnish roots?

Yeah..smooth move on my part.. rofl.. I 'prep' my VR's with little 4" leads on them..so its easier to get them mixed up with my poor organizational skills, I guess.. :shrug:

I have removed two SMR's for a VID increase to 1.2xx volts..I'll post some pics later so you can let me know what I goofed up on. :ROTF: ..just got in from work..its 3am here.. *sleep*

and ya.. with a last name liek "Juntunen" its kinda hard to deny the heritage [not that I was trying] :D

Hopefully tomorrow, I can swing by my electronics supply and grab the right VR's this time, and get something going on these GPU's finally. I will be modding 8x of them, for use in a folding farm for a member over on OCN.

*ATT's keeping me 'booked' lately with work all over the state..so trying to put time in on the mod when I can..

[whodie]

Well CL3P20 kicked me in the ass yesterday....so it's time for me to get back into the "real" world


I've been going about modifying vGPU via the BIOS, rather than VR, but my results have been trying my patience

1.1v is the normal BIOS vGPU limit - I've made several attempts to modify the voltage tables and artificially increase the VID but varied success. My latest attempt was setting 3D VID at 1.23, Extra at 1.26 - it's been a disaster, although I'm not 100% sure what the root of the problem is yet. Core temps increase 7C (from 1.1v) - 3 of the 4 GPU's (used for folding) run fine at tame speeds of 700/1836/900, but the 4th crashes within 10 minutes. At 725/1890/950 2 of the cards run fine, the other two crash almost instantly.

The problem is that is a nearly impossible task for me to measure vGPU on all four cards while they are installed and I simply don't have the time to test/measure each one individually at the moment :( My gut tells me though that the VID specified in the BIOS isnt being correctly applied - even at 1.1v all 4 cards could do 700/1758/950.....


Now, what I'm curious is if I remove the resistors AND set the VID to ~1.3v will it actually work and not induce OVP?

[CryptiK]

Changing VID will not trigger OVP. It shifts the OVP point up accordingly. See largon's post:


VID determines OVP trip point which according to the datasheet is typically VID+170mV so setting VID->1.31250v will never trigger OVP.

[whodie]

Agreed - but my concern is setting VID at 1.3v and removing the resistors

With those resistors removed and VID set at 1.3v wouldn't the actual vGPU be higher than 1.3v? Certainly the variance would be higher than 170mV if default VID is 1.1v and removing the resistors shifts this to 1.3v :shrug:

[largon]

As stated, "removing the resistors" means you're setting a new VID, and as such a new OVP trigger point. Thus, whatever you do to the VID resistors you will not see OVP kicking in.

[whodie]

Sorry - I should have been more clear - when I stated setting VID to 1.3v I meant via the BIOS

I was referring to altering the BIOS voltage identifiers to push VID to 1.3v in conjunction with physically removing the resistors. Hopefully that makes more sense! lol

In all honesty I'm probably fighting a losing battle trying to adjust VID in the BIOS :shakes:

[quadomatic]

Awesome, I got my multimeter and VR's today. Sadly I don't really have the time to get at this (still in high school...2nd semester senior so not much really needs to get done but still, got a project to do...), and I still need to pick up some 30 AWG wire. I ran out and all I have is 22 AWG, and I think 22 is a bit big for this.

And using air cooling, an Accelero S1 with the turbo module, which now runs in the mid 50 degrees range on load with my current oc, how much could I likely raise the voltage to? Sadly I don't use water cooling, but still.

Also, what ground do I use for measuring voltage while the card is on? Is it any of the back three pins of the PCI-Express connector? Should i be fearful of shock?

UPDATE: I played around with one of the VR's, measuring resistance. So, turning clockwise while the VR is connected by the right 2 pins causes resistance to decrease, increasing voltage, while turning counter-clockwise increases resistance and decreases voltage.

Or, connecting by the left 2 pins and turning clockwise causes resistance to increase and voltage to decrease, while turning counter-clockwise decreases resistance and increases votage.

You guys probably already knew that, but I'm a n00b so this is interesting for me :P

I have it set to maximum resistance right now. I'm not sure if my multimeter is inaccurate or whether its just the +/- 5% error but I read 98.4k ohms as the resistance on my 100kohm vr.

Does the system have to be off while resistance is adjusted?

[CL3P20]

1ST thing... YES the system needs to be OFF to measure resistance values.. measuring while powered = bridged resistor = BOOM.. no resistor :D

..as for ground to use while measuring GPUv ..I just ground to my case or tech station.. Its generally not a good habit to probe with ground and positive leads, while powered. Also.. for the VR's.. you can just clip one of the outer legs off, if you dont wish to solder both [as shown in my pic, on previous page].

*I would say, 1.3v idle should be OK for air cooling.. shader clocks will benefit from having cooler core temps though..which might be attainable easier [on air] at a lower GPUv.

...just my :2cents: ..Im sure others will have things to add. Good modding.

**Hey Whodie !! ..birds of a feather..:D

[quadomatic]

Thanks for the response, but I'm confused by a couple things. First, I know that I can't measure resistance while it's on, but could I adjust the VR without measuring resistance and just measuring voltage all while the system is powered on?

Also:



..as for ground to use while measuring GPUv ..I just ground to my case or tech station..


so do you just put the red probe on the GPUv point and the black probe on the metal of the case? That works?...I guess I could try it right now. Again, do I need to worry about shocking myself or something? I'm a little paranoid...

BTW: Any results on your mod?

UPDATE: I tried measuring the voltage. I measured 1.18V using the upper vGPU point, which seems odd since it should be 1.1V I guess. I tried reaching for the deeper point to see but it was too far away to reach. Anyways, after I did that I guess I must have poked somewhere I shouldn't have because my display got all distorted with white dots all over it like it was crosshatched like crazy. I rebooted and its all good now I guess.

UPDATE 2: Tried measuring voltage again. I didn't miss this time, and my computer didn't die afterwards. I measured using both points and used my case as the ground, and I got 1.18V. I was warned that the multimeter I bought wasn't the best one around though, so it's very possible that it's just off on it's reading. Somewhere I read that voltage goes up ~ .05V in 3D. I'm using Aero. I might've read that in this thread...but I can't find where...oh yeah it was in the thread Largon wrote on how to volt mod 8800 series cards.

I did try to pencil mod the card unsucessfully. I tried to erase as much as I could, but since that "mod" didn't give me any results I'm guessing that maybe the card just runs at this voltage. I could try disabling Aero and measuring again...maybe later.

[CL3P20]

Fixed*
adjust the VR and just measure voltage while the system is powered on? :D

Your findings are pretty normal.. most of the 8xxx and 9xxx GPU's I have mod'd all reacted the same way.. with GPUv increasing on load ~.04v.

*As for the soldered mod.. I havent made it to the store yet.. hoping to get over there tomorrow.. busy w/ the kids an all

[largon]

I know that I can't measure resistance while it's on, but could I adjust the VR without measuring resistance and just measuring voltage all while the system is powered on?Yes, once you have the VR soldered correctly you can adjust it even when the card is powered while monitoring the voltage.

[quadomatic]

OMG....This is not good:

http://img442.imageshack.us/img442/8549/img1907fj9.th.jpg (http://img442.imageshack.us/my.php?image=img1907fj9.jpg)

http://img383.imageshack.us/img383/9426/img1911to1.th.jpg (http://img383.imageshack.us/my.php?image=img1911to1.jpg)

Foolishly, I tried to do the vMEM mod. Unfortunately, I lifted the trace, and ripped off the fb pin in the process...I don't suppose there's anything I can do to fix this...? Or would one of you guys be able to fix it?

[CL3P20]

omg.. I am sorry for your loss. :(

Short of getting a whole new chip..and attempting to replace the whole thing.. other than that.. you might be able to 'jump' the connection with a short wire..soldered in place to the connecting resistor..

*I imagine Largon, to be more experienced in this department... Repair is not my specialty :o ..usually I smash broken parts and keep the mangled pieces as trophies.

[quadomatic]

omg.. I am sorry for your loss. :(

Short of getting a whole new chip..and attempting to replace the whole thing.. other than that.. you might be able to 'jump' the connection with a short wire..soldered in place to the connecting resistor..

*I imagine Largon, to be more experienced in this department... Repair is not my specialty :o ..usually I smash broken parts and keep the mangled pieces as trophies.

I was thinking of jumping the connection. I need to see if I can actually get a wire to attach to the chip though...there's still a TINY bit of the leg of the chip showing, so if I got a glob of solder to sit on it, it maybe could be done. Or maybe I could take a knife/dremel to it to make it show more of the leg.

Or maybe it's just not worth the trouble and I should pick up a new video card....gaaahhh!!! So much money...sure this card was just $34, but still. I'm not going to find another card like this for that price...

[quadomatic]

I took the blade of a scissor and started scraping at the chip to expose the leg:

http://img150.imageshack.us/img150/9041/img1913pa8.th.jpg (http://img150.imageshack.us/my.php?image=img1913pa8.jpg)

I guess I exposed something. I think I'll try soldering to it and then soldering to the resistor...omg I'm scared.

[quadomatic]

Tried soldering to it, and solder won't adhere. Now I can't even see the exposed leg anymore. I guess this card is pretty much done for...DAMNIT.

Looks like I'm going to have to look for something new. I bet someone will buy it on eBay.

But if anyone has suggestions on what I can do, please let me know.

Maybe get a HD4830 and pencil mod it lol...

[CL3P20]

GPUv mod confirmed!! Pics of the ordeal below.. THANK YOU LARGON !! :D



pic of mod.. with correct VR: 100k ohm
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050551.jpg

reading resistance: post mod
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050554.jpg

corresponding voltage, to above resistance level
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050553.jpg


..well I got 1x down... 7 more to go now :D

[shoehorned]

Sorry for being lazy and not reading through the 4 pages of technical stuff, but has vmod successfully worked for anyone here?

[CL3P20]

how did you miss the post above yours..? :rolleyes:

[shoehorned]

But has it translated into higher clocks in the card?

[CL3P20]

I havent finished testing the GPU's yet for OC.. but increased GPUv = higher clocks.. that is never really in question. Previous efforts at this mod, have not resulted in increased GPUv.... if that answers your Q any.. if not, I'll guess youll have to wait for screenies of the OC, wont you?!

[largon]

CL3P20,
Please test if there is an aggressive OVP. Tune it to, say, 1.40v to see if/when it trips.

quadomatic,
Ouch!
=(

There are several pin-compatible chips (RT9259, ISL6549, eg.) that are very common on different motherboards and video cards, if you have some dead components laying around you might want to consider scavenging for such a part and attempting to replace the chip.

[CL3P20]

CL3P20,
Please test if there is an aggressive OVP. Tune it to, say, 1.40v to see if/when it trips.

.. Roger that.. testing today. Will post back some screenies, and pics of the 'mini mod-farm'.

[quadomatic]

CL3P20,
Please test if there is an aggressive OVP. Tune it to, say, 1.40v to see if/when it trips.

quadomatic,
Ouch!
=(

There are several pin-compatible chips (RT9259, ISL6549, eg.) that are very common on different motherboards and video cards, if you have some dead components laying around you might want to consider scavenging for such a part and attempting to replace the chip.

I decided it's not worth the trouble. If the card had cost me $100 and it performed very well, I'd go for it. But, since the card only cost me $34 and it's probably not going to last TOO much longer as far as it's performance compared to the current generation of cards, and since it's apparent my soldering skills are sub-par at best, I've decided to throw it on eBay and let somebody else use it for parts or try to fix it themselves.

Still though, how often do people rip legs and lift traces like that? It's probably because I was yanking the wire to remove it when the joint was a bad cold joint. I guess it was still good enough a joint to tear :banana::banana::banana::banana: up. Lesson learned.

I think I'll pick up an HD4830. A simple bios flash vmod can let you oc it to past the speed of an overclocked 8800GT. Also I'll still have the Accelero S1 and the turbo module I bought for the 9600GSO. It should fit on the HD4830, and then I'd get better clocks than I've seen from others. At $90, it's the best I'm going to do. I don't really want to spend too much, since I'm going to college this year and it seems likely that I won't be able to take my computer with me.

[CL3P20]

Some updated pics and runs ..+ mod farm pics, of the 7x GPU's I just finished...



..In the process of finishing up 7x [8th GPU completed already and not shown] mod'd EVGA Dual-slot 9600GSO's [96sp] for a client..

These will be running air cooled with GPUv mods for daily folding use. Preliminary results are good.. getting ~720mhz core and 1898mhz shaders @ ~1.15v idle.

Here are some pics of the fiasco under way..


GPU's fresh out of the EVGA store packaging..
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050565.jpg



backside of GPU's..
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050568.jpg



coolers removed for modding..
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050570.jpg



overview of the 'mod station' ..aka meh work bench :D
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050566.jpg



one of the core shots.. very nice batch #.. similar to GS/SC editions
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050573.jpg




..shot with GPUv mod completed
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050591.jpg



setting ~3.68k ohms on the VR = ~1.18v idle GPUv
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050554.jpg

..the 7x GSO's shown, should net ~35k PPD together.. should be wonderful addition to the clients farm.


AS5 going on..
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050592.jpg

Glue drying on vmeasure points
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050593.jpg

completed 9600GSO with GPUv and v-measure point [relocated for easy access]
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050596.jpg

the test rig..
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050600.jpg

monitoring GPUv under load..
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050599.jpg
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050605.jpg
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050607.jpg

Screenies as promised :D

quick bench run with 3D 06, using 1.288v

792mhz core
1826mhz shader strap
1038mhz mem
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/modd_foldingclocks_gpu8.jpg

with these settings for folding use, you would gain ~1200 PPD over stock.

http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050621.jpg
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050611.jpg

...also, for those dual-slot 9600 owners.. Zalman RAM sinks DO fit under the cooler, on the mosfets.. so no worries about keeping them cool anymore. :thumb:
http://i265.photobucket.com/albums/ii220/CL3P20/GPU%20mods/P1050614.jpg


Sorry for the late RE... and @ largon- I havent had any time to check for OVP limits yet.. sorry.. been busy though. :D

[quadomatic]

Very nice! Not a bad core OC. Try pushing shaders more?

I found out I'll be getting a 9800GT relatively soon, so I'll be okay.

Don't think I'll be modding anytime soon though...

[CL3P20]

@ quadromatic- Thanks! ...stock cooling is preventing better clocks, thats all... obvious in the shaders case.. I cannot hold them at the 1898mhz strap, for stable use with load temps over ~55c.

[quadomatic]

@ quadromatic- Thanks! ...stock cooling is preventing better clocks, thats all... obvious in the shaders case.. I cannot hold them at the 1898mhz strap, for stable use with load temps over ~55c.

Really? Can't be over 55? On stock I was still stable on 60 something.

What oc's were u getting before you increased voltage?

[CL3P20]

before mods, most all the GPU's were topping out around 630mhz core and 1650mhz shaders... after mods, setting the shaders for 1889mhz develops shader arties within a few minutes of benching... lowering them to the previous strap setting, of 1836mhz cleared all artifacts.

*these are similar results to my GS/SC's.. keeping temps below ~50 on load, allowed me to run 2166mhz for shaders for stable use.. Keep in mind the GPU's I am using came from EVGA store...refurbed.. so I think results are pretty good, considering.

[quadomatic]

Ah, from eVGA store. I guess I might've gotten really lucky with my card then, considering I got much better OC's on stock voltage.

[OrangeGator]

I did the vMem mod and it works great. :clap: Before the mod my max stable speed was about 930mhz and vMem was 1.9V. After, with the resistor still set to 10K and using the same ground point CL3P20 used in his pics, I got 2.0V and am at 1060mhz and still testing.

I'll do the vGPU once I pick up a 100K vr. I'm a little confused about raising VID by removing the 2 resistors that largon showed. If I removed those resistors, would the voltage for vGPU increase to 1.3V from the current 1.18V without adding the 100K vr? Or would just the over voltage cutoff threshold be raised, but the vGPU remain at 1.18V (unless the 100K vr is used)? Thanks for the great info!

edit: I got impatient and removed those two resistors as suggested by largon hoping it'd raise vgpu without using a vr. Bad Idea! They came off with little problem, but vGPU has dropped to 1.06V, from 1.18V previous! My previous overclock of 660/1674 is nolonger stable. But stock speed still is. I'm going to pick up a 100K VR and see if I can fix this.

[largon]

OrangeGator,
The VID-mod changes vGPU without the VR.

Please see/measure if the pads on which VID4-resistor was mounted on are still connected. Like via a tiny bridge of solder.
That would explain your voltage drop as VID4 makes quite a difference:
1.31250V is VID0: shorted (0), VID1: no connection (1), VID2: no connection (1), VID3: shorted (0), VID4: no connection (1), VID5: shorted (0) and VID6: no connection (1).
And VID for 0.93750V is the same as 1.31250V, but 0.93750V has VID4 shorted (0).

[OrangeGator]

OrangeGator,
The VID-mod changes vGPU without the VR.

Please see/measure if the pads on which VID4-resistor was mounted on are still connected. Like via a tiny bridge of solder.
That would explain your voltage drop as VID4 makes quite a difference:
1.31250V is VID0: shorted (0), VID1: no connection (1), VID2: no connection (1), VID3: shorted (0), VID4: no connection (1), VID5: shorted (0) and VID6: no connection (1).
And VID for 0.93750V is the same as 1.31250V, but 0.93750V has VID4 shorted (0).

Hey largon. Are you sure that shorted =0, open =1? It seems like it's the other way around, as quadomatic guessed earlier. But I'm a noob and really don't know. I'm almost certain that the resistors were removed and there was no solder bridge.

Anyway, long story short, I fixed it. I added a 100K vr and have tuned vGPU to 1.31V. Currently stable at 753/1890 and still testing. No problem with overvoltage protection so far. My advice, do Not remove resistors. I have a feeling that the diagram showing which vid resistor is which is incorrect or removing them doesn't have the results we'd expect.

Ok, long story for those interested. After removing the two resistors and getting a vGPU of 1.04, I was able to successfully re-solder vid resistor 4 back and got 1.16V. I then tried to re-solder vid 2 back, and remove vid 3. Well, I removed vid 3, but for the life of me could not get vid 2 back on. Damn things are small and wouldn't stick. In the process, vid 4 came off and would not go back. So, in desperation, I put solder bridges over vid 2 and vid 4, leaving vid 3 open. This resulted in 1.14V. I then picked up a 100K vr today and added the mod. I was then able to tune vGPU to 1.31V from the 1.14 previously.

[largon]

Copy-paste from RT8802A datasheet referring to the VID tables:
Note: (1) 0 : Connected to GND
(2) 1 : Open

[quadomatic]

Good to see more progress has been made!

Unfortunately for me, I don't seem to be getting much of any performance gains out of the 9800GT. I can't get the core, shaders, or memory as high as it was on the 9600GSO either.

[CL3P20]

not too unusual.. more onboard cache with the extra 16 SP's = more internal heat to be removed, before instability occurs.. Better cooling is in order.

[quadomatic]

not too unusual.. more onboard cache with the extra 16 SP's = more internal heat to be removed, before instability occurs.. Better cooling is in order.

Actually it's probably not a heat issue. I replaced the reference model cooler with an Accelero S1 and a Twin Turbo module. Temps peak at 55 degrees or so. It's probably voltage related, but I'm not so inclined to do a volt mod anymore :P

Maybe a bios flash. It might not be set to 1.15V.

[OrangeGator]

Good to see more progress has been made!

Unfortunately for me, I don't seem to be getting much of any performance gains out of the 9800GT. I can't get the core, shaders, or memory as high as it was on the 9600GSO either.

I'm happy with the final results. My OC went from 660/1674/930 to 756/1836/1080 stable @ 2.0V vMem, 1.32V vGPU. Temps top out at 72C with stock cooler. I tried higher volts, but couldn't get anything higher stable due to temperatures. With better cooling, I'm sure it'd clock even higher. But as it is, I get 12,566 3dmark06 with my e7200 @3.7ghz.