Request: GPU Memory Junction Temperature via nvidia-smi or NVML API

The lack of this feature made it unnecessarily difficult to debug a thermal/power consumption issue on my Thinkpad P1 Gen 6’s RTX 4000 Ada. I had to deduce that the reason for the “SW throttling” was actually the VRAM overheating, modify a reverse-engineered utility just for it to be able to read my gpu’s vram temperature, then confirm with Lenovo’s engineers if the behavior was correct.

Here’s the public part of said discussion:
https://forums.lenovo.com/t5/Fedora/Thinkpad-P1-Gen-6-No-thermal-response-for-rising-VRAM-temperatures-until-t-junction-max-reached/m-p/5279930

There’s no excuse for this value not being exposed when it’s already been proven that userspace on Linux is able to read it. I shouldn’t need to mess with libpci just to track a temperature sensor’s value on my workstation graphics card.