9600GT voltmod thread

[largon]

Technically on-the-fly is possible but it's not really about what the IC is capable of but the possible VID combinations in the table and the exact order the switches MUST be flicked to prevent unwanted VID-code changes. It's not that simple as to just flick them according from VID=1.15V to VID=whatever without paying attention to how the VID code is changed by each flick of the switch... There's a 400 nanosecond time window during which the switches must in the intended position, or else the VID currently set is applied. So one has to follow an exact - often multi-staged - path from a given voltage to another so that VID-code related drops & spikes are avoided. One wrong move and *whoops!* -> 0.7500V -> reset.

[wittekakker]

Ah indeed, I've always made sure to shut down the machine first just avoid stuff like that happening.

[fatbird2046]

Still cann't break 1.4v wall!
I have paralleled the 150k reistor with 100k,and replaced the 170k one with another 100k.But the wall is still there!



I still think it is because of PWM switch frequency setting or Current limit,so i will keep searching for it.

[trez0r]

What and where should i solder to get 1.3v?

[largon]

trez0r,
To get 1.3V remove these bridges:

[Xargo]

trez0r,
To get 1.3V remove these bridges:

Shorted = 0

For 1.3V you need to unshort VID4 and short VID5.

[trez0r]

Thanks a lot largon!
I will be back with results..

[largon]

Shorted = 0

For 1.3V you need to unshort VID4 and short VID5.

No, shorted is 1.
The bridge chips are 0Ω "resistors".

[trez0r]

Actually Xargo is right (0 = shorted). I tried the other way around and couldn't even boot in win - volts were much below 1v.

Anyways no luck with this mod for me - i set it to 1.3V (which is 1.35v under load) - and guess what!? Again i get black screens when doing atitool artifact scan... Bah.

[Xargo]

Actually Xargo is right (0 = shorted). I tried the other way around and couldn't even boot in win - volts were much below 1v.

Anyways no luck with this mod for me - i set it to 1.3V (which is 1.35v under load) - and guess what!? Again i get black screens when doing atitool artifact scan... Bah.

Yep, 1.2V seems to work best for me thus far. I hope someone will find a way around the ocp or whatever is restricting the overclock.

I'm pretty pleased with 13k score but it's still a bit frustrating because there clearly is more potential in this chip.

Largon,

Yes, I know they are 0Ω "resistors" but that doesn't change the fact that shorting results in 0. Easy enough to check with a multimeter and if you compare the original "resistor" placement with the vid-chart you'll notice that vid5 is the only one not shorted (1) resulting in 1.1V (0100000).

[largon]

Easy enough to check with a multimeter and if you compare the original "resistor" placement with the vid-chart you'll notice that vid5 is the only one not shorted (1) resulting in 1.1V (0100000).

Hmm, you are correct.
I actually thought 0 was open and 1 closed pads because of a sequence of coincidences starting from the pic in this post. I see that VIDs 2, 3 and 4 are closed and think that stock vGPU is 1.15v (= 0011100).
But infact, stock vGPU isn't 1.15v but 1.10v.

I guess this would be easier for me if I actually had a 9600GT...

[Xargo]

I guess this would be easier for me if I actually had a 9600GT...

Yep, I figured as much =). But you are correct in that the default vgpu (for 3d) is 1.15V, but in 2d-mode it's 1.1V (and the table tells you the 2d-mode voltage). So my 1.2V config is actually 1.25V in 3d-mode and it seems to be plenty enough for me now. I agree that this layout is a bit awkward though.

This is a nice card but I probably step-up when some other 9000-series are introduced (I have evga board), unless someone can sort out the ocp or whatever is limiting the clocks here.

[mike]

Couple of things that might help - only played with the card for a couple of hours but it seems that the - XP driver suks - Vista driver will be much more stable ( and no more crashes in ATItool etc. ). Higher Vcore appears to hurt shader clock - keep the shader clocks lower and you might be able to get the GPU clocks up.
Curreently running @ 890/1850/2400 voltages 1.25V (1.3V 3D loaded ) / 2.08V
Not possible for me in XP
Shaders I was able to run 2300mhz with default Vcore. Might be a fluke - but you guys might find similar behaviors

[trez0r]

Yeah, i could run core at 1000+ with lowered shaders with atitool artifact scan @1.3v, but whats the point? Shaders are the bottleneck of this card anyway.

Wish someone will figure out the workaround of black screen problem. I think my card is a real beast, too bad i cannot fully explore the potential of it yet.

[trez0r]

Still cann't break 1.4v wall!
I have paralleled the 150k reistor with 100k,and replaced the 170k one with another 100k.But the wall is still there!



I still think it is because of PWM switch frequency setting or Current limit,so i will keep searching for it.

fatbird2046:
What do resistors mentioned in picture do? Do they raise the card's tolerance to higher vgpu?

[fatbird2046]

fatbird2046:
What do resistors mentioned in picture do? Do they raise the card's tolerance to higher vgpu?

No they don't raise the tolerance to higher VGPU.That is what OVP mod should do.

But the trouble is at 1.4v the card can run for a few seconds when running 3Dmark2006 then system will reset.

If it is OVP,you cann't even restart the system with this 1.4v voltage.

But if the PWM frenquency or current limit has been set to low lever by manufacturer.When the load is not so heavy(starting XP and just running 2D Program),the card is ok,when the load becomes heavy(like running 3dmark2006),the current will be much higer than in default voltage,and need higer PWM frequency to transport more power.So if the default current limit or PWM frequency setting is low,the controlling chip will restart the system.

In ADP3207,these 2 parameters can be set by changing 2 resistors' value,this 2 resistor is all connected to ground.

But the problem is we don't have ADP3208's datasheet, so we have to guess and try....

[trez0r]

fatbird: i tried paralelling the resistors - didnt help me at all - card still shutdowns when voltage is raised and load is applied in atitool.

[kaptan54]

Couple of things that might help - only played with the card for a couple of hours but it seems that the - XP driver suks - Vista driver will be much more stable ( and no more crashes in ATItool etc. ). Higher Vcore appears to hurt shader clock - keep the shader clocks lower and you might be able to get the GPU clocks up.
Curreently running @ 890/1850/2400 voltages 1.25V (1.3V 3D loaded ) / 2.08V
Not possible for me in XP
Shaders I was able to run 2300mhz with default Vcore. Might be a fluke - but you guys might find similar behaviors

nice clocks.
has any benchmark score?

[LyP0]

so.... still no vmod ?
If not than what's the point of putting it under ln2 :@
Was gonna try that when my cpu arrives but this doesnt look good.
maybe i'll return the card

[Zeus]

so.... still no vmod ?
If not than what's the point of putting it under ln2 :@
Was gonna try that when my cpu arrives but this doesnt look good.
maybe i'll return the card

You give up rather soon, check out Hicookie at OCX forums, he's running 1.4Vgpu without problems on a 9600GT.

[largon]

so.... still no vmod ?

See posts #22 and #24:

vGPU mod is pin #6 of ADP3208 to ground (through a 100kΩ VR).

vMEM mod is pin #5 of "51117" to ground (through a 200kΩ VR).


Attachment 72993
Attachment 72994

[Fr3ak]

largon, is that mod working for sure? I think I will give it a go tomorrow and see what happens. Together with Shamino, we should be able to figure the OVP mod out.

[largon]

Yes, the mod works alright.
According to the datasheet of ADP3207, disabling OVP should be very simple: pin #35 (on ADP3208A) to ground using just a jumpwire.

[trez0r]

You give up rather soon, check out Hicookie at OCX forums, he's running 1.4Vgpu without problems on a 9600GT.

Is he running reference card or is it with custom pcb w/ 3 phases?

Eh, i'm still patiently waiting for what is i guess OCP mod..

EDIT: On the other hand maybe there is no workaround on this black screen problem and maybe it's just because of weak power phases on reference design. Tho hope dies last.

[largon]

trez0r,
You got an OCP-related PM.