TUI with chart for radeon power cap control, temp monitoring

amdgpu-power-control.py

#!/usr/bin/env python3
import plotext as plt
from subprocess import check_output
import psutil
import time

CMD = "sensors amdgpu-pci-0500 | grep PPT | grep -oP '(?<=cap =) *(.?\d+)'"
TARGET_TEMP = 83
CRITICAL_TEMP = 92
CAP_MIN = 2
CAP_MAX = 180
CHART_HISTORY_COUNT = 55
plt.theme("pro")

# Prefill arrays for good chart after start
a_temp = [43]
a_powercap = [180]
temp = psutil.sensors_temperatures()["amdgpu"][0].current
powercap = int(check_output(CMD, shell = True).decode('utf-8'))
for i in range(1, CHART_HISTORY_COUNT):
        a_temp.append(temp)
        a_powercap.append(powercap)

while True:
        # Trim chart
        if len(a_temp) > CHART_HISTORY_COUNT:
                a_temp = a_temp[1:]
                a_powercap = a_powercap[1:]
        # Get data
        temp = psutil.sensors_temperatures()["amdgpu"][0].current
        a_temp.append(temp)
        powercap = int(check_output(CMD, shell = True).decode('utf-8'))
        a_powercap.append(powercap)
        # Plot temp
        plt.clear_terminal()
        plt.clear_data()
        label = "Temp %s°C" % temp
        plt.plot(a_temp, label = label, color = "red")
        plt.ylabel(label)
        plt.ylim(20, 100)
        plt.hline(TARGET_TEMP, "green")
        # Plot power cap
        label = "Powercap %sW" % powercap
        plt.plot(a_powercap, label = label, yside = "right", color = "blue")
        plt.ylabel(label, yside = "right")
        plt.ylim(CAP_MIN - 10, CAP_MAX + 10, yside = "right")
        plt.show()
        # Hard save
        if temp >= CRITICAL_TEMP:
                # Kill all non-root(avoid suicide) python processes
                check_output('ps -ef | grep "[p]ython" | awk \'$1!="root" {print $2}\' | xargs -r kill -9', shell = True)
        # Limit power
        newcap = powercap
        if temp < TARGET_TEMP:
                cap_step = TARGET_TEMP - temp
                newcap = int(min(powercap + cap_step, CAP_MAX))
        if temp > TARGET_TEMP:
                cap_step = pow(temp - TARGET_TEMP, 2)
                newcap = int(max(powercap - cap_step, CAP_MIN))
        if newcap != powercap:
                check_output("/opt/rocm/bin/rocm-smi --setpoweroverdrive " + str(newcap), shell = True)

        time.sleep(1)