(Potential) Fix for "pwr" limit of MXM GPU

[JadeRover]

Hello eveyrone, 

 

Ok so this problem applies only to MXM gpus that have a "pwr" limit (as well as "thrm" limit sometimes) in GPU-Z when doing a benchmark. This is also known as the "40w power bug" Such example :

 

 

Notice this is not due to hitting the power limit of the card (33% of TDP).

 

Symptoms of the low power bug is a TDP usage around 50% = 40w if you have a TDP of 80w.

 

I think I found the issue. It has to due with a digital signal that is sent to the GPU, the PWR_LEVEL signal :

 

 

A previous fix was to block the SMB_DAT or SMB_CLK signal so that the EC chip of the laptop wouldn't be able to communicate with the MXM dGPU. This has the disadvantage of stopping all data communication between EC and MXM dGPU = no more thermal data coming in = EC usually ramps up the GPU fan to maximum.

 

According to MXM 3.1 documentation :

 

 

This signal is not mendatory and it does apply some kind of power reduction when enabled. Therefore I took a gamble and decided to cover it up and see what happens.

 

My fix is to block off the PWR_LEVEL pin with a bit of tape. 

- I first put some normal paper scotch tape on my plastic desk (to get it nice and flat during cutting process)

- I cut it into a thin strip with a very sharp cutter(width of MXM pin)

- I removed it from my plastic desk, then applied it with tweezers.

 

The pin is the 9th counting from the first "thin" pin, on the TOP of the MXM module aka side of the MXM gpu where there is the actual GPU die :

 

It's okay if you cover up the 8-5th pin as well as these are reserved pins (no connection inside a laptop)

 

 

I did this on my dell precision 7720 with GTX 1060M and this has so far resolved the issue, my card can now boost to 100w thanks to the modded bios. 

 

Hope this helps someone !

[sliderfra]

Hi,

That's great job! Too bad I sold my alienware 17 already, I would've tested it with the RTX 3000, as the TDP bug happened on it.