gpu-temp-monitor: an optimum clocking tool for AMD GPUs

I got inspired by this other thread (http://forum.bitcoin.org/index.php?topic=10062.msg167868#msg167868) and wrote a bash script to manage GPU core clocks versus temperature. gpu-temp-monitor supports up to 8 AMD GPUs (or more, by editing the initialization of the "holds" array).

The script assumes that you have DISPLAY and whatnot set up correctly in your environment, and that you've opened up a nice range for over/underclocking with AMDoverdrivectrl or similar.

Code:

#!/bin/bash

TEMP_HIGH=95 # degrees C
FREQ_MAX=900 # MHz
FREQ_STEP_UP=5 # MHz
FREQ_STEP_DOWN=30 # MHz
MEM_FREQ=150 # MHz
HOLD_TIME=15 # sec
TOLERANCE=2 # degrees C
IGNORE_TEMP=$((TEMP_HIGH-TOLERANCE))
OUTPUT_INTERVAL=1 # sec

holds=( 0 0 0 0 0 0 0 0 )

setclock() {
	id=$1
	clock=$2
	if [ $clock -gt $FREQ_MAX ]; then
		clock=$FREQ_MAX
	fi
	aticonfig --adapter=$id --odsc=$clock,$MEM_FREQ 2>&1 > /dev/null
	holds[$id]=$HOLD_TIME
}

reclock() {
	id=$1
	temp=$2
	clock=$3
	flag='-'
	hold=${holds[$id]}
	if [ $hold -gt 0 ]; then
		holds[$i]=$((${holds[$id]}-OUTPUT_INTERVAL))
	elif [ $temp -ge $TEMP_HIGH ]; then
		setclock $id $((clock-FREQ_STEP_DOWN))
		flag='v'
	elif [ $clock -lt $FREQ_MAX -a $temp -lt $IGNORE_TEMP ]; then
		setclock $id $((clock+FREQ_STEP_UP))
		flag='^'
	fi
	echo -n "$temp ($clock$flag) "
}

aticonfig --od-enable 2>&1 > /dev/null

while true
do
	unset T
	unset C
	temps=`aticonfig --adapter=all --odgt | grep : | sed 's/.* - \([0-9]*\)\..*/\1/'`
	clocks=`aticonfig --adapter=all --odgc | grep "Current Peak" | sed 's/.*: *\([0-9]*\) .*/\1/'`
	i=0
	for val in $temps; do
		T[$i]=$val
		i=$((i+1))
	done
	
	i=0
	for val in $clocks; do
		C[$i]=$val
		i=$((i+1))
	done

	i=0
	while [ $i -lt ${#T[*]} ]; do
		reclock $i ${T[$i]} ${C[$i]}
		i=$((i+1))
	done
	
	echo
	sleep $OUTPUT_INTERVAL
done

And the mandatory megascreenshot: