T61 Nvidia GPU undervolting: success!

This is my re-post from thinkpads.com

Recently I had finished a project on building a T61 15” Flexview Frankendpad.
The donor T61 is 8889-ACG and unfortunately it has dedicated NVS-140M GPU. Comparing to Intel X3100 integrated video it has some serious drawbacks:
- It is prone to Nvidia GPU bug with bumping material.
- It consumes a lot of power: at idle my T61 FV consumes 14.3W with all off, backlight at lowest level, and Powermize is on max battery. While another T61 15.4” WXGA with X3100 GPU consumes only 9.5-10W. Take into account that my Flexview UXGA requires 1-1.5W more than the 15.4” WXGA, I estimated that the Nvidia GPU consumes about 4-5W at idle, while the Intel one consumes about 1-1.5W.

So my biggest concern is to lowering power consumption of Nvidia GPU. Using Nvidia Inspector and NVPMManagerUni, I set the GPU at fixed max battery level and underclock the GPU a bit, to 130Mhz. However, the power consumption was still high because the supply voltage for GPU was still 1.15v. Therefore, my next objective is to lowering the GPU voltage.

However, unlike CPU voltage, the mobile GPU voltage can not altered by software. Some PC video card can change voltage by altering the so called voltage table in its BIOS using NiBitor software, but this software does not recognize mobile video card where video BIOS is integrated with main BIOS. Therefore, the only way I can go with is to make hardware mod in the GPU power supply system.

Studying T61 schematics, on page 85 I had found that Lenovo uses ADP3209 IC to control the voltage of GPU. This IC has 5 VID pins (like CPU power supply systems). Before going into action, I tried to overclock my GPU, and it could go as high as 650Mhz at 1.15V. So my estimation is that it can keep the default frequency of 400Mhz at about 0.90V. Based on that, I decided to give the VID3 voltage level of 1 instead of 0.

To do this mod, I removed the 0 Ω resistor R590 from the board to give high (1) signal level to pin 26 (VID3) of the ADP3209 IC. However, due to unknown at this moment reason, the voltage at this pin is still low (0). To cope with it, and due to the fact that before I accidentally remove R537, I connected pin 26 to pin 27 (VID2) and they both have high level. The GPU voltage after this mod is 0.95V for Powermize lowest performance level (before: 1.15V) and 0.90V for highest performance level (before 1.20V). You may find strange that the latter voltage is lower than the former

The laptop boots without problems, at 0.90V I even can overclock the GPU a bit, up to 450Mhz. Using Video Stability test, I had found that the GPU temperature droped by 8-10 degrees Celcius (from 74deg to 66 deg according to GPU-Z, or from 65 to 58 deg according to TPFan control). Another effect is that at idle, the laptop consumes only 13.0W instead of 14.3W before the mod. Thus, the battery life on my 9 cells 77Wh battery is extended by 10%, up to 4.5 hours comparing to 4 hours before. Hope that with lower working temperature, my Nvidia GPU will work longer

Here are images of ADP 3209's VID voltage table and part of T61 schematics. The actual picture of mod will be posted later.

 

Impressive work. You can see the heathy voltage margin provided to the gpu. Usually CPUs are also overvolted beyond their needs. Your technical expertise means you could consider a companion guide to overclock the cpu based on the info in Lenovo/IBM T61 FSLx PLL pinmod.

The dual-IDA + PLL pinmod would extend the longevity of these great systems. Your T9300-2.5/2.7 would be running at 3.33/3.59Ghz, which is comparable to i5 cpu levels.

 

Hmmm, I don't quite understand how you modded your hardware to lower the GPU voltage, but congratulations on the mod! From What I have seen, GPU voltages can be easily modified with Nibitor with many notebooks, but I suppose there are always exceptions, in which case you have to find other way around.

I find it odd why you give more voltages to the lowest performance level than the highest level, if the GPU remains stable at 0.9v when stressed, why give more than 0.9 to the lowest level?

 

I don't have much motivation to overclock CPU due to the 2 facts:
- I tried to do PLL pinmod on my previous T60p but the laptop did not even boot. My guess that the BIOS on T6x series does not allow for CPU FSB frequency change. Note that X-series are overclocked easily while there no precedence of overclocked T60 or T61.
- My T9300 at 2.7Ghz requires 1.1V to work stable (while at 2.5Ghz requires 1.0125V). My calculation shows that at 2.9Ghz it will require about 1.25V. It is too high and my cooling system will hardly cope with it.

 

Nibitor does not allow for altering voltage table of my BIOS (I used Phoenix BIOS editor to decompose Lenovo BIOS and then open it in NiBitor). Moreover, I think that playing with BIOS without specialized facilities is a very dangerous business.

The voltage in minimum performance level is higher than that in maximum performance level can be explained in the updated VID table in my first post. Before the mod, VID2 = 0 for max perf. and =1 for min perf. I set VID3=1 to lowering voltage by 0.2v, VID2=1 to lowering it by another 0.1v. But in min perf. mode VID2 already =1 so it has no effect, i.e. only VID3 has effect.
Another way to explain: please, look at the value(s) of VID1 in that table.

 

congrats.

I run mine at 625MHz core and 950 memory. For over 3 years I have had no problems, thus no intentions of actually undervolting the GPU.

the max it can run is at 650/975

I think the 140m on the 15" thinkpads were using 256MB memory. Maybe those were the ones that would hit the dust often. Mine is a 128MB version, no issues so far.

 

I'm trying to overvolt the 7150M on this laptop and I don't understand how to do it with the voltage table and all that? I keep getting an error about them not being correct. On 3d it says 1.2V. I'd like to up my 3d volts to 1.25-1.3V... I don't understand how to do it.

I'm having issues with my tl-60 overclock when at 2.5Ghz from 2.0Ghz it runs cool under intelburntest at 68*C and ocasionally the "cpu" temp will spike to 98*C for a split second. The Cores are still at 68-69*C when that spike occurs. Maybe it's the voltage regulators overheating? Anyway it'll pass sometimes and sometimes it'll fail. I suppose 2.427Ghz may be the stable max however in game 2.5Ghz runs great. I'm using GPU-Z to pull video bios, and CrystalCPUID to adjust cpu voltage. At 2.0Ghz it uses 1.150V I believe. I pushed up to 1.325V and it didn't break 75*C and wasn't stable. However in game it's stable. I like to pass intelburntest though. Anyway I need more gpu clock. Maybe 1.35V on the gpu core to push to 650Mhz. The 7150M has a hard time with heavy geometry.

I wanna overvolt the gpu since when i overclock to 575Mhz from 425Mhz it becomes unstable but it's only at 63-65*C when at 575Mhz. I need to over volt the bios so I can push it to 600Mhz stable hopefully and still stay below 70*C on the GPU.

I modded the laptop bottom plastic to let the fan breathe and it dropped temps about 30*C since the fan was restricted from lack of design.

*UPDATE*

Installed copper shim on cpu and i'm only hitting 58*C on intelburntest with the 500Mhz overclock totally stable. I'm running the 7150M at 560Mhz core clock totally stable and running pretty damn good i must say! Kills an X3100 by far! Just need to overvolt so I can run 650Mhz+. At 560MHz core clock I'm seeing 73*C in 3dmark03 on the top end. I could see it going to 75-78*C with an overvolt and overclock to 650Mhz without a problem. The 7150M isn't that bad at all!

I'm thinking about getting a TL-68 2.4Ghz for $75 and push it to 3Ghz hehe.

7150M - 3dmark03 - Battle Of Proxycon - 15fps avg
GTX 470 - 3dmark03 - Battle Of Proxycon - 5,000fps avg lol

 

great work, always wondered if this could be done.

 

I successfully overclock my T61. Now I have a T8300 running stable at [email protected].
Long story. I will report about it later when have enough time.

 

so you pulled down that pin on the clock gen to start using 266MHz bus instead of the 200MHz one ?

12x 266 would be 3.2GHz, as far as I can see the things, heh.

nice

 

I don't think I'd wanna push my T8300 even stock on intel burntest for 20 runs the core run about 75*C but cpu reports 85*C. That would be cool to push past 2.4Ghz but I'm running the GM965 intel chipset so it's locked down.

I found where my stability issue was with the Turion 64 x2 was my memory was at ddr2-883 from dd2-667. So now I'm ordering some ddr2-800 ram to run stable at 2.5Ghz on my TL-60 2Ghz Turion 64 x2 chip. I wonder if the turion 64 x2 at 2.5Ghz would beat the 2.4Ghz T8300 core 2 duo.

 

No. My method is a bit different (from those developed by Chinese guys). I pulled up to VCC the BSEL1 line on the chipset side, and pulled down this line on the clock generator side. Moreover, I made a pull-down switch to make the overclock "switchable". After pulling up BSEL1 on the chipset side, you can do whatever you want like regular BSEL mod. All of this thing can do without soldering / cracking motherboard. It is safe in T61 because there is a 1K resistor on BSEL1 line between chipset and clock generator.

Also I made dual voltmod: both overvolt and undervolt. Now my voltage rage is between 0.770v and 1.330v (with some unavailable voltages) instead of original 0.925-1.225v

You can overclock on 965GM chipset with some extra effort with GPU (there is successful case, but I don't remember correctly).
My CPU temperature is around 80deg Celsius running Prime at [email protected].

 

Can you help me find the right vid to get an undervolt on my 2d clocks and an overvolt for my 3d clocks?

http://forum.notebookreview.com/gam...-can-you-help-me-dial-vid-voltages-7150m.html

 

what clock gen chip do you have in your T61 ?

I've opened a 14" T61 wide that was here before and there was a SLG8LP564V clock gen chip, and on the bottom side there was no resistor there on pin 7. I haven't traced the top side though.

post some pictures when you have time

the issue with mine now is that if I do the same thing then I'll have to do overvolt as well since the clock is going to jump to 13x 266 = 3.45Ghz which I'm pretty sure would not run at the stock VID of 1.137 Volts. I'm thinking that it would want something like ~1.3 Volts, which at full load would heat up quite a lot.

heck I've tried 2.8GHz at 1.2250V and even with TPFanControl running the fan on max, temps went towards 90+ deg C in couple minutes, so I said "screw that" and went back down to 1.0750 V

cant let this thing to heat a lot due to my 50%+ overclocked GPU ... hehe.

good job on your side though

---
P.S.
Thinking further of that, T8300 has only 3Mb cache so will keep cooler at same speed compared to T9300/T9500 with its 6Mb cache. Post some benchmark of your CPU at 3.2GHz to see how it actually performs. You can use the Intel LinX loading program for that, it produces results in GFlops. Google it and I'm sure you'll find where to download it from. As far as I remember, mine at 2.8GHz produced on average somewhere around 17.5 GFlops, so waiting for your results

If it scores higher then T8300 would be the new choice of CPU for OC enthusiasts, as may not require further VID increase mods ... LOL.

cheers.

EDIT:
here's what VID somebody on here used to run his X9000 at:

3,0GHz - 15x - 1,1375 V
3,2GHz - 16x - 1,2125 V

so past 3.2GHz the voltage increase becomes quite high.

 

so I've calculated estimated performance based on the CPU performance charts from here:

PassMark CPU Benchmarks - High Mid Range CPU's

T8300 - stock score 1,489 => OC to 3.2GHz is 33.33% increase => estimated score 1,985

T9500 - stock score 1,839 => Dual IDA to 2.8GHz is 7.7% increase => estimated score 1,980


P.S.
- obviously X9000 at 3.2 GHz smokes both at estimated score 2,262, which is 23% over stock T9500

EDIT:
- the IDA mode of the CPUs probably bump the scores by a little here and there for the stock scores, so my calculations could be a little off. X9000 does not have IDA.

 

OMG .. I'm so wanting to try a X9100 CPU right now ... LOL.

Lock the chipset at 200MHz, feed 266MHz to the CPU, and get that thing up in speed by changing the multiplier with ThrottleStop

- sorry for the triple post, but I got excited as it seems

 

OK. Tomorrow I will try to describe about all of these hacks (at least in a short version).
To cool you down a bit here are some more information:
- The biggest problem when overclocking is IDA mode. You can disable it in Middle_tone's BIOS (Power -> SpeedStep: Disable). However, I have a feeling that this not fully disables IDA.
- Because of IDA, T9300 and T9500 are very difficult to OC. I tried T8300, T9300, T9500 (all have good voltage at stock speed). T9500 has a light crash on one core at 3.2Ghz @ 1.25v (I suspect to IDA mode, because it takes only 1.065v at 2.8Ghz). T9300 can be stable at 3.2Ghz but at higher voltage, and due to bigger cache, it is hotter than T8300 (up to 90 deg. C). So I stick with T8300.

Send me e-mail address and I will send to you something useful for hacking your T61.

 

As promised, here is my report on overlocking / voltmoding my poor T61

 

@ tqbinh

Where did you get the schematic? I need a schematic for the NVIDIA MCP67/ NVIDIA GeForce 7150M so I can see if I can get more than 1.2V in 3d since the the hex voltage table in nibitor shows that there is nothing above 1.2V. I just wanna get 600-650Mhz out of the core clock from it's 425Mhz stock clock and I can reach 550-560Mhz on stock voltage but 600-650 would be great and my temps are fine about 60C in game.

http://www.mediafire.com/?63smfp56m15trt9

 

amazing work. Love my t61p, but this project seems like it'd take too much time.

 

I am somewhat confused by that paragraph:

1. Wouldn't removing the resistor break the pathway of electric impulse? Shouldn't it be replaced with some kind of non-resistant connector?

2. Instead of removing the resistor, would it work if I just pencil it over with a soft graphite pencil? I used to do it back when I was using desktop systems and it worked well. I am not really good with a soldering iron... nor have one readily available, for that matter.

3. This accidental removal of R537 - are we supposed to replicate it and remove it, too, after which we apply the fix proposed in this description? Or should we just leave it be AND apply the fix? Or should we just ignore that part, as we never actually removed R537 in the first place?

 

here I am resuscitating an old post! but given that a T61 never gets old... : -)

I really would like to attempt this mod, I think that reducing the Nvidia temperatures of 8-10 degrees makes total sense! (that's the only reason why I miss the X3100... it was so much cooler & quieter!)
I already have an undervolted T8300 cpu (the coolest cpu apart from T8100), so the Nvidia is constantly the hottest part, even at idle.

from Binh's post I understand that I need to remove R590 and R537 resistors (are they resistors?) and connect pin 26 and 27...

sounds fairly doable, even for someone without experience like me... my problem is: how can I individuate where these resistors and pins are located on the motherboard..??? I'm totally lost there.. maybe I need the schematics in order to individuate those?

help much appreciated, even in PM if necessary : -)