This is a continuation of my other older thread located here:
http://www.linuxquestions.org/questi...-a-4175544458/

I am just posting my experiences to date with Slackware 14.1 and kernels 4.0.4 and 4.0.5 on a Dell Latitude D620 as other might find it useful if they have the same machine.

Since my last posting in the other thread I ended up switching back to run level 3 as I found out that the screen would blank out black with only the mouse cursor present on the screen after X had been running for a while. Sometimes it was not possible to even get X to start up at all. Instead only the black screen with the mouse cursor was all I could get. When this happens, there is no way to shut the machine down gracefully. Although it is possible to log in from another computer with ssh, it wasn't possible to shut the machine down gracefully. The shutdown would just hang. I eventually discovered this screen blanking was related to the ghost images I was seeing during boot up. If I saw the ghost images then I knew not to start up X an if I didn't see them, I could start X up with no problem. I eventually learned that the ghost images were actually artifacts from an over heating GPU. and not being caused by either of the kernels frame buffers I was running. The GPU on this laptop is the NVIDIA Quadro NVS 110M, which is know to have this problem as I ended up finding out. :-(

After examining the heat sink on this laptop, I came to the conclusion that a bad cooling system design was the real reason for the GPU over heating and not the GPU itself. There are three chips that the heat sink is trying to cool down. I presume one of them is the bridge chip, while the other two are the processor and the GPU. The processor has the advantage as it is in front of the other two chips heat sink wise. The GPU and what I presume to be the bridge chip are on separate branches of the heat sink that connect to the main part of the heat sink were the processor is. With this type of a configuration, the heat sink cooling capacity is being exceeded and so the GPU either ends up slowly cooking itself or shutting down to prevent that kind of damage. I immediately got Linux-PHC installed and under volt the processor to help keep the processor cool which in turn has helped to keep the GPU much cooler than it was. Before under volting, the GPU would get as high as 162 F (72 C) and eventually shut down. The lowest would be 142 F (61 C) at a minimal load. After under volting, the GPU runs between 129 F (54 C) and 158 F (70 C). It only gets that hot when I am compiling something for a couple of hours or more. When I watch a video, it only gets to 142 F (61 C). I stuck the under volting setting in my /etc/rc.d/rc.S file and right at the beginning so the processor will get under volted as soon as possible. The main thing is, that the GPU is no longer over heating and the artifacts at boot up have disappeared. So I have switched back to run level 4. I am now also able to run an opengl screen saver as before the under volting the screen saver would cause the screen to blank out immediately.

The latest Intel release of Linux PHC works fine with kernel 4.0.4 but has an invalid symbol error at load time under kernel 4.0.5. So for the time being under volting the processor is not possible under kernel 4.0.5. However the GPU doesn't appear to be over heating under kernel 4.0.5 either. My research indicates that the Nouveau driver has the ability to change the GPU's clock speed just like the NVIDIA driver. So my guess is that the Nouveau driver might not be setting the clock speed under kernel 4.0.5 and the GPU is running at the lowest clock speed and so not over heating as a result. I'll have to get with the Linux PHC community on this problem.

I was also looking into the possibility to under volt the GPU under Linux. It doesn't look like that is a possibility under Linux. Although it does appear to be a possibility under Windows with certain GPU's. The closest thing I could find was to use NVIDIA's Coolbits software to change the clock speed as a possible means to keep the GPU cool. Which means trying to get NVIDIA's driver to work. I haven't had much luck with it so far. I'll either need to hunt down the patches to the NVIDIA driver or else try the latest version of the driver that has just recently released.

I had this machine a while back. It is known to have a faulty NVidia GPU. Somewhere in the process of manufacturing these Dell got new chips from NVidia.

If you have an NVIDIA chip where the letters "Nvidia" are filled in white, it's the old (faulty GPU). If the letters are hollow, you have a new GPU which shouldn't have the thermal problems anymore.

I threw mine away when the GPU broke because a new mainboard was too expensive for me.

You can try to do a copper mod, where the heatpad on the GPU is replaced with copper shims, that seems to work on other devices.

I also have an Amilo here with a Go 6800. It has the same thermal problems and is known to fail.

I underclocked it with nvlock: http://slackbuilds.org/repository/14.1/misc/nvclock/ - now it runs cool.

You can undervolt Nvidia GPU's by writing a new ROM to it. There is a good windows tool for it, called Nibitor:
http://www.guru3d.com/files-tags/(nibitor).html

You can lower the voltages, write the new Rom and reboot. But be careful, if the voltage is too low you won't get an image anymore.

I don't know for sure, but I probably have one of the older ones.

I might consider doing this sometime in the future.

This requires NVIDIA's driver too.

I think this might be the best way to go. If the voltage is set too low, does this affect the text displays too?

Strange. On my GF6800 I can set the -f switch and it forces the new clocks by writing directly to the registers.
It warns about the unorthodox approach, but it's doing it

If the voltage is set too low you can get all kinds of graphical glitches, up to the device not running at all anymore. I think it's really dangerous to do that on notebooks, unless you have a USB graphics card to recover later.

I have the same laptop with the nvidia heat problems. I copper shimmed the gpu with
new silicon with no good results. The only cure I have found is "i8kfan" and just
run the fan at a higher speed.

I tried nvclock 8.4b and it didn't do anything. Maybe you're using a later version of the Nouveau driver. I'm just using what came with Slackware 14.1. I see nvclock has the ability to use Coolbits as a backend, so it will work with the NVIDIA driver.

I'll probably just set the voltages down gradually and see what temperature I get with the GPU. I'll try not to go any lower than necesary. I finally tracked down a dos version of nvflash, so hopefully I'll get this going soon. Nibitor runs under Wine so I'll do the rom file edit there and then flash it with nvflash under dos. I'll also be sure to keep a copy of the original rom file just in case the edited rom file has a problem flashing.

I found a console script that throttles the processor if it's temperature gets to a certain level. https://github.com/Sepero/temp-throttle I changed it to monitor the GPU temperature instead. It has helped keep my GPU temperature from getting any hotter than 158 F (70 C) during long compiles. I start it up at boot time by adding this to the end of /etc/rc.d/rc.S

/usr/local/bin is where I put the script. Nohup does create a nohup.txt file under / where screen output goes.

I have dellfand running in the background, however the bios seems to be fighting it for control of the fan. Anyway dellfand sets the fan to a high speed and the bios sets it back to low. The end result is that the fan ends up pulsing at a high speed for about two seconds. I am using a modified version of dellfand that I made that allows for optional Farenheit temperatures and another temperature setting to keep the fan from pulsing at high speed. It works great on my Dell Latitude D400 but not so great here. I have never had much luck in getting I8K to work in the background so I use dellfand. I don't know if I8K would have the same problem or not. I just wish there was a way to turn off the bios fan control permanently. I do know of a way to turn it off temporarily from the thermal control panel but it only lasts a few minutes.