voltmod G73M / Geforce Go 7600? (Amilo Xa 1527)

This is on XTB70 motherboard, used in both of the FIC Everex XT5000T & Fujitsu Amilo Xa 1527 [ Link for schematic on Scribd]

Skip to "The Question" section further below if you don't want to read a lot.

Background Info:
At idle, and at the lowest performance state (100/135Mhz), GPU typically runs 48C-52C hot, after having replaced the cheap thermal pads on GPU & NB & Heat-pipe-junction with copper shims + thermal paste, from original temps in high 50's / low 60's.
Because this chip is vulnerable to the NVidia bug (which I've baked on griddle & reflowed with KADA hotair + liquid-rosin-flux + thermocouple monitoring, to resurrect it about 2.5 months ago [link for Detals]), I have made some BIOS mods in an attempt to mitigate the issue:
1. SSDT table MOD to undervolt CPU. [ Link for Details]
2. DSDT table MOD to spin fan slightly faster for idle and also at lower active cooling level + added fan full speed for top active cooling level + making AC0/AC1 levels activate sooner by adjusting temperature trigger points / hysteresis levels. [ Link for Details]

Extracting the vBIOS from either laptop's main (Phoenix) BIOS and opening it in NiBiTor shows that VID Bitmask is "01", and so only VIDs "00" (1.1V) and "01" (1V) are available. Default was 1V for 2D & Throttle states, and 1.1V for 3D.
I modded the vBIOS so that 3D state uses 1V instead of default 1.1V, to drop load temps from 85-88C to 77-78C. [ Link for Details]

The Question:
I would like to get both idle/load temps down lower by undervolting further, which means I need a hardware volt-mod.

Page 57 of the XTB70 schematic shows the VID control logic (VDD_SEL) but I'm unsure how to proceed to achieve desired undervolting of both VIDs; say to 0.8V/0.9V, or 0.85V/0.95V.

I know it would be quite difficult to voltmod on-board mobile GPU but I think I can tackle it, if I could just get some pointers on what I need to measure and/or modify. Any help in this regard would be greatly appreciated.

Cheers

 

Shouldn't be too hard if you have some soldering skills. I increased the voltage of my 6970m a while ago, check the link for details.
You want to achieve an undervolt, so it's practically the same, but instead of lowering the resistance between vsen and ground you'll need to increase it.
So you can do almost the same as I did, but you'll need to put the potentiometer serial between vsen and ground, not parallel.

 

No ISL and no pin marked vsen in the VGA_VDD page. Have a "SC413TSTRT TSSOP-28". I cannot find a datasheet for it (only SC413, but no TSSOP28). Though pin 12 seems relevant, it is marked FB2. I assume that's short for Feedback?
Though it works for you, the method you describe seems more like a workaround, with some guesswork involved. Wouldn't directly modifying the VID control (and VID labels in vBios), provided a schematic (as in my case) be more "scientific", and therefore less risky?
Thanks

 

ISL is just a manufacturer, and vsen is just a label for a pin, don't expect them to be the same for your mobo. I only linked to the 6970m so that you can see how it is done. This method can be applied to all GPUs that use such a kind of VR.
It's pretty much impossible to say anything certain without having the datasheet. There's no guesswork involved in my method at all. I'm simply using some of the controlling circuits of the VR to work according to my needs.
Of course you can easily change the VID labels, but as the name says, they're only labels. GPU-Z will report a lower voltage but it won't affect the voltage of your GPU. Your GPU only supports two voltages, which are set by the VR on the board. You'll need to do a hardmod to achieve an undervolt.

 

Sorry. I understand what a PWM chip is, manufacturers, labels, etc.
What I meant by guesswork is, after setting VR to adjust the resistance between the sense/feedback pin and ground, you have to measure what the vCore changed to, until you reach a desirable vCore. i.e. You're not measuring to double-check a calculated value against actual result.

Now, you had issues with voltage getting setback - when using furmark. Have you considered the fact that 3D state (in vBios) uses a different VID (higher voltage) than 2D/Throttle states. i.e. Does your MOD account for the vCore changes (changing both performance state voltages to a desirable value), which is controlled by the VID selection logic on the board?
From what I see, you only have one measurement (14Ω .

I've looked at the semtech SC415 pdf (the closest match to the SC413 I could find) and now pretty certain that FB1/FB2 are the feedback pins, used for adjusting the VOUT1/VOUT2 voltages, and to which the VDD_SEL circuit is connected.
My current thinking was to calculatingly understand how this VDD_SEL circuit sets 1.0/1.1V, and modify it directly to achieve desired voltage. Am I even making any sense?

 

Hi Marco,

To clarify on what I meant by:

I got a bit more help and figured it out here:
Help undervolt-mod G73M / Geforce Go 7600 (+ Mobo Schematic link)

Cheers to you

 

So per this followup post, I concluded that increasing the resistor between the FB pin and ground (R55) will give more predictable results than decreasing the resistor in the FB path (R60). This makes a lot of sense because R55 is in the VDD_SEL control logic circuit. This also happens to be akin to your method doh me!

Now that I fully understand the voltage selection logic, the need for a potentiometer is obviated. That is, I can calculate to determine the replacement needed for R55 to achieve the exact desired volt-mod.

And as you said: I'd have needed to put the potentiometer in series between FB & GND, not parallel. This would involve breaking the circuit somehow, most likely by lifting/replacing R55, to put the potentiometer in series. So it's the same pain anyway; have to lift R55 off the board regardless of using a pot or replacement resistor. Though, I reckon that using a replacement resistor would be more preferred over a pot (for a permanent mod). Right?

 

My voltage issues were/are caused by the insufficient power supply of the card. As you probably realized, the mod I suggested is independent of the power states of the GPU, it raises or respectively decreases the output voltage independently of the pin(s) which the GPU chooses as voltage source.

I can't look at the schematic atm, but usually the easiest way to put something serial is to simply cut the conducting path carefully with a knife or similar, then you add a resistor to the path, serial. I recommend using a small trim potentiometer since it'll allow you to easily adjust the GPU to the desired level (you want to undervolt, there's no way to know in advance which min voltage the chip needs to be functional). So it'll be easier to correct if it is too low/high. And it will fit inside the notebook.

Also you probably now this, but I repeat it anyway... be careful when working with a multimeter. Measuring the resistance between two points usually is not a good idea as long as the system is running...

 

MOD Details:
Removed R55 (10KΩ , soldered thin wire to the positive end of its place in the circuit, and soldered a wire to ground (the DC jack housing), then connect these to wiper and a terminal of a 20K pot (adjusted to 10K). Also soldered a (red) wire to the positive end of C15, to measure vCore (in reference to ground).

Findings:
With stock vBios, VIDs correlate to 1 and 1.1V
However, with R55 exactly = 10.00KΩ, The readings are a tad higher, 1.025-1.028V and 1.128-1.130V, respectively. That's about 0.028- 0.030 overvolt, in the stock configuration.

Now, from extensive testing, the lowest vCore that will run 2D perfectly is 0.945V, by adjusting R55's replacement pot to 11.93KΩ. That makes the 3D voltage = 1.048V. So that's ~0.08V drop in both 2D and 3D voltages, from stock.

By flashing a modded BIOS (vBIOS module modded) so that 3D uses same VID as 2D, I was able to determine that lowest vCore possible for 3D without corruption = 0.987V.

But, I wanna keep the lower 2D voltage (0.945V) while having ~1V for 3D.

If I only modify R60 (leaving R55=10KΩ , that wouldn't get me much closer:
R1 = (Vout/FB - 1)*R2
R60 = (0.945V/0.5V - 1)*10 = 8.9KΩ
when VDD_SEL=L,
Vout = 0.5V (1 + 8.9/8.333333) = 1.034V

So, best option is leaving R55=11.93K, and modding R44 to adjust the 3D voltage.

To get Vout = 1.0V:
R2 = 10/(1.0V/0.5V - 1) = 10KΩ
And, Since:
R2 = 1/(1/R55 + 1/R44), when VDD_SEL=L
Therefore:
R44 = 1/(1/R2 - 1/R55)
And given new R55 = 11.93KΩ
R44 = 1/(1/10 - 1/11.93) = 61.8KΩ

So I guess one could connect another 20K pot in series between R44(50KΩ and Q8 to add ~ 11.8KΩ to effectively drop the 3D voltage to ~1.0V

But I decide it is not worth the effort to do any further MOD. As is, with R55 @ 11.93KΩ, I'm getting 1.048V for 3D, and the temp difference for another -0.05V drop will probably be negligible. Besides, this should give me a little head-room to up the 3D clock a bit, for fun.

Pics and illustrated steps of the mod here: http://hardforum.com/showthread.php?p=1038565164&posted=1#post1038565164