#!/bin/bash
# This script runs in the background and checks the temp of my 3060 and 3090 cards.
# The concept is to cut voltage when approaching a thermal limit. This "smoothes out"
# the usage compared to native thermal throttling. This is useful for long sessions
# of Stable Diffusion or heavy chat models. You may need to tune for your own setup,
# temperature preferences, and so on. This is provided with ZERO WARRANTY, so your
# mileage may vary. In general, you should not do magic you do not understand.

p0max=170
p1max=370
tmax=72
tmin=55

let p0=$p0max
let p1=$p1max
sudo nvidia-smi -i 0 -pl $p0
sudo nvidia-smi -i 1 -pl $p1
let s=0

while [ true ]
do
    clear
    nvidia-smi

    zzz=10
    d0=`nvidia-smi --query-gpu=temperature.gpu,gpu_name --format=csv,noheader | grep 3060`
    t0="${d0:0:2}"
    echo 3060 Temperature $t0. Present Watts $p0. Max Watts $p0max
    p0action="3060 (=)"

    if [[ $t0 -gt $tmax ]]
    then
        let p0=p0-1
        sudo nvidia-smi -i 0 -pl $p0
        zzz=1
	p0action="3060 (-)"
	s=0
    fi

    if [[ $t0 -le $tmin ]] && [[ $p0 -lt $p0max ]]
    then
        let p0=p0+1
        sudo nvidia-smi -i 0 -pl $p0
        zzz=2
	p0action="3060 (+)"
	s=0
    fi

    d1=`nvidia-smi --query-gpu=temperature.gpu,gpu_name --format=csv,noheader | grep 3090`
    t1="${d1:0:2}"
    echo 3090 Temperature $t1. Present Watts $p1. Max Watts $p1max
    p1action="3090 (=)"

    if [[ $t1 -gt $tmax ]]
    then
        let p1=p1-1
        sudo nvidia-smi -i 1 -pl $p1
        zzz=1
	p1action="3090 (-)"
	s=0
    fi

    if [[ $t1 -le $tmin ]] && [[ $p1 -lt $p1max ]]
    then
        let p1=p1+1
        sudo nvidia-smi -i 1 -pl $p1
        zzz=2
	p1action="3090 (+)"
	s=0
    fi

    if [[ $s -gt 10 ]]
    then
        s=0
	if [[ $p0 -lt $p0max ]]
	then
	    p0=p0+1
	    sudo nvidia-smi -i 0 -pl $p0
        fi
        if [[ $p1 -lt $p1max ]]
	then
	    p1=p1+1
            sudo nvidia-smi -i 1 -pl $p1
        fi
    fi	

    echo $p0action $p1action
    sensors gigabyte_wmi-virtual-0
    sleep $zzz

done