XPS 15 9570 Overheating GPU

> when I run "sensors" in Linux, nearly all temps are close to 100C. 

I believe that direct CPU / GPU throttling is built-in (can't be disabled) and should cause clock downthrottling rather than shutdown normally. Disabling Intel DPTF does not do away with all throttling. Your condition must be really bad. 

Have you checked if your cooling fins are clogged with dust? If not, it might be really bad thermal paste, poorly attached heatsink, or a cracked heatpipe. 

When your laptop is already open, also do place some thermal pads on the VRM mosfets, because they badly lack active cooling and are known to overheat, sometimes even blow up. 

Thanks for the tips!

I have concluded that it's not the GPU, It seems to be CPUs that run hot, I replaced the thermal paste (The old was really bad), fixed thermal pads on other chips then also disabled "turbo boost" on CPU in BIOS. Now it seems to work fine. GPU was never that warm.
I'll see if I can find some cooling for the mosfets too before I enable turbo boost again.
I do not think I had any dust problem, I looked OK and I cleaned them just a month ago.

I still think this is a bios issue or something, but I not to keen on downgrading and fiddle with bios versions.

> I have concluded that it's not the GPU, It seems to be CPUs that run hot

> GPU was never that warm.

Mind the thermal limit of the GPU is much lower than the 100 degC of the CPU, in the seventies, and GPU throttling tends to hit fps hard in games. See using HWinfo64\Sensors

> fixed thermal pads on other chips 

Stock VRAM pads appear to be okay, at least there was no report of improvement by changing them - though one or two beside the fan may not make contact, because the heatpipe is bent there to adjust the level difference between the chips and the fins. 

> I'll see if I can find some cooling for the mosfets 

Thermal pads are easy to apply - but they mostly prevent local overheating, while taking much heat away isn't a simple task (check iunlock's mod). Presumably for the 9570 the pad sheet "bridge" to the heatpipes (not touching the backplate) is preferred - if you take heat to the backplate it may induce throttling from the DIMM sensor sooner. 

>  disabled "turbo boost" on CPU in BIOS 

This is a good idea for gaming, to help avoid GPU throttling, while the CPU performance hit doesn't tend to matter so much. For other purposes it may not be desired. You can change CPU power level on-the-fly using Throttlestop or such (not sure for linux). 

After doing my below I could get the computer to run OK:

- In bios set thunderbolt enumeration to "bios assisted enumeration" (It switched randomly in auto mode, and native did not work for windows, screen flickered like crazy. linux I'm not sure how this setting worked or not).

- In bios disabled "turbo boost". Without it both CPU and GPU overheated to 100C

I think there is some error in bios or other chip FW after 1.13.0 bios/FW upgrade.

Problem is still there, even without turbo boost disabled, It just takes longer time for it to overheat.

I think thermal throttling is broken.

The problem is still there, I sent the results from "self test" with thermistor values" to support, they said they look normal.

So then, there is another evidence that there may be som issue in BIOS och thermal throtteling code in processor micro code?

Laptop still overheats if I use "turbo boost" in bios _OR_ use really high resolution in games. 

Hi,
Thanks for writing in. The temperatures are fine as per the picture shared earlier. Is the system heating when the turbo boost is disabled?
Regards,
Mausam

Hi,
I hope you're doing well. I am awaiting your response to earlier messages. I will go ahead and archive this case for now, as I haven't heard from you. Feel free to reach out if you have any concerns. Here is the reference number for our conversation:42916289

Regards,
Mausam

Hi,
We did not receive a response from your end.
The system is built in a way that it can sustain temperatures to 100 degrees.
Is the system only heating while gaming?
Which application are you using to measure the temperature?

Regards,
Mausam

Yes, but not as much as when it's enabled.

Hi,
I have replied to your private message.

I tried to rollback BIOS. I didn't want to go too far back so I've set it to 1.9.1 version. Still occurring.

Temperatures for CPU were around 90°C and GPU at 83°C when it rebooted this time. CPU Turbo is disabled. Applied heavy load on both GPU and CPU for testing purposes.

I'm having the exact same issue. Having all sensor data displayed right before shutdown I can confirm that it shuts down without any apparent sign of overheat. Same BIOS log saying thermal shutdown event has occured.

I have monitored CPU frequencies and there was no throttling and the system suddenly shuts down when using GPU.

I also upgraded BIOS recently to 1.13 and now 1.14, the problem is still there.

Also I tried both Nvidia 430 drivers and 440. Same issue, sudden poweroff.

I've booted to Linux kernel 5.3.15-1 instead of usual 5.4.2-1 today. I did a heavy charge test using ~100% GPU and ~100% CPU (unigine-valley + dummy dd), temperatures stay under control and CPU throttles accordingly. No sudden shutdown any more. It appears the problem originates either from kernel 5.4 or Nvidia drivers for this kernel "linux54-nvidia-4X0".

Problem solved for me, but I hope I'll be able to upgrade my kernel at some point.

As a side note: when it works, throttling is violent btw, sudden drop of CPU frequency from 3GHz to 800Mhz, not very smart, CPU turbo mode will be disabled for me from now on...

Another side note: the emergency procedure for shutting down in case of overheat is neither safe nor fast enough. It actually triggers the power button, which means it has to wait for the same delay that applies when the user presses it (~12 seconds) before turning off the machine. This is way too long if the processor temperature has exceeded 100°C, lot of room for a Chernobyl incident inside your computer.