Script to verify cpu thermal throttling

check_thermals.sh

#!/bin/bash

# --- CONFIGURATION ---
DURATION=60          # How long to run the stress test (in seconds)
SAFE_TEMP_LIMIT=98   # Emergency cutoff temperature (Celsius)
PASTE_SPIKE_THRESHOLD=25 # If temp jumps this much in 2 seconds, suspect bad paste
THROTTLE_THRESHOLD_PCT=85 # If CPU speed drops below 85% of max, suspect throttling

# Colors
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
CYAN='\033[0;36m'
NC='\033[0m' # No Color

echo -e "${CYAN}=== CPU Thermal Health & Throttling Detector (Arch Linux) ===${NC}"

# 1. Check Dependencies
if ! command -v stress &> /dev/null || ! command -v sensors &> /dev/null || ! command -v bc &> /dev/null; then
    echo -e "${RED}Error: Missing dependencies.${NC}"
    echo "Please run: sudo pacman -S stress lm_sensors bc"
    exit 1
fi

# 2. Get CPU Info
CPU_MODEL=$(grep -m1 "model name" /proc/cpuinfo | cut -d: -f2 | xargs)
MAX_FREQ=$(lscpu | grep "CPU max MHz" | awk '{print $4}' | cut -d. -f1)
# Fallback for Max Freq if lscpu fails
if [ -z "$MAX_FREQ" ]; then
    MAX_FREQ=$(cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq)
    MAX_FREQ=$((MAX_FREQ / 1000))
fi

echo -e "CPU Model: $CPU_MODEL"
echo -e "Max Rated Speed: ${YELLOW}${MAX_FREQ} MHz${NC}"
echo "------------------------------------------------------------"

# Function to get current max core temp
get_temp() {
    sensors | grep -E 'Core|Package|Tctl' | awk '{print $3}' | grep -o '[0-9.]*' | sort -nr | head -n1 | cut -d. -f1
}

# Function to get current avg CPU freq
get_freq() {
    grep "MHz" /proc/cpuinfo | awk '{sum+=$4} END {print sum/NR}' | cut -d. -f1
}

# 3. Phase 1: Idle Check
echo -e "${CYAN}Phase 1: Measuring Idle Baseline (5 seconds)...${NC}"
sleep 2
IDLE_TEMP=$(get_temp)
echo -e "Idle Temperature: ${GREEN}${IDLE_TEMP}°C${NC}"

if [ "$IDLE_TEMP" -gt 55 ]; then
    echo -e "${YELLOW}WARNING: High idle temp detected! (Possible bad airflow or background load)${NC}"
fi

# 4. Phase 2: Stress Test
echo "------------------------------------------------------------"
echo -e "${CYAN}Phase 2: Starting Stress Test (Load for ${DURATION}s)...${NC}"
echo "Use Ctrl+C to abort immediately."

# Run stress in background
stress --cpu $(nproc) --timeout $DURATION &
STRESS_PID=$!

# Monitoring Loop
START_TIME=$(date +%s)
SPIKE_DETECTED=false
THROTTLING_DETECTED=false
MAX_TEMP_REACHED=0

# Check initial spike after 2 seconds
sleep 2
CURRENT_TEMP=$(get_temp)
TEMP_RISE=$((CURRENT_TEMP - IDLE_TEMP))

if [ "$TEMP_RISE" -ge "$PASTE_SPIKE_THRESHOLD" ]; then
    SPIKE_DETECTED=true
    echo -e "${RED}ALERT: Instant Temperature Spike Detected! (+${TEMP_RISE}°C in 2s)${NC}"
fi

while ps -p $STRESS_PID > /dev/null; do
    CUR_TEMP=$(get_temp)
    CUR_FREQ=$(get_freq)
    
    # Update Max Temp
    if [ "$CUR_TEMP" -gt "$MAX_TEMP_REACHED" ]; then
        MAX_TEMP_REACHED=$CUR_TEMP
    fi

    # Safety Cutoff
    if [ "$CUR_TEMP" -ge "$SAFE_TEMP_LIMIT" ]; then
        echo -e "${RED}EMERGENCY: Temperature reached ${CUR_TEMP}°C. Stopping test to prevent damage.${NC}"
        kill $STRESS_PID
        wait $STRESS_PID 2>/dev/null
        break
    fi

    # Check Throttling (If temp is high AND freq drops below threshold)
    FREQ_THRESHOLD=$(echo "$MAX_FREQ * $THROTTLE_THRESHOLD_PCT / 100" | bc)
    if [ "$CUR_TEMP" -gt 85 ] && [ "$CUR_FREQ" -lt "$FREQ_THRESHOLD" ]; then
        THROTTLING_DETECTED=true
        echo -e "${RED}>> THROTTLING DETECTED: Temp: ${CUR_TEMP}°C | Freq dropped to: ${CUR_FREQ} MHz${NC}"
    else
        echo -ne "Running... Temp: ${CUR_TEMP}°C | Freq: ${CUR_FREQ} MHz   \r"
    fi
    
    sleep 2
done

echo -e "\n------------------------------------------------------------"
echo -e "${CYAN}=== DIAGNOSTIC REPORT ===${NC}"

# 5. Analysis
echo "Max Temp Reached: ${MAX_TEMP_REACHED}°C"

# Diagnosis: Bad Paste
if [ "$SPIKE_DETECTED" = true ]; then
    echo -e "${RED}[BAD RESULT] Thermal Paste/Contact Issue:${NC}"
    echo "   - Your temperature spiked instantly (+${TEMP_RISE}°C) when load started."
    echo "   - This usually means heat cannot move from the CPU to the cooler."
    echo "   - ACTION: Reseat cooler and re-apply thermal paste."
elif [ "$MAX_TEMP_REACHED" -gt 90 ]; then
    echo -e "${YELLOW}[WARNING] General Overheating:${NC}"
    echo "   - Temps are very high, but the rise was gradual."
    echo "   - ACTION: Clean dust filters and check fan curves."
else
    echo -e "${GREEN}[GOOD RESULT] Cooler seems functional.${NC}"
fi

echo ""

# Diagnosis: Throttling
if [ "$THROTTLING_DETECTED" = true ]; then
    echo -e "${RED}[BAD RESULT] Thermal Throttling Confirmed:${NC}"
    echo "   - Your CPU slowed down to cool itself."
    echo "   - Performance is being lost due to heat."
else
    echo -e "${GREEN}[GOOD RESULT] No Throttling Detected.${NC}"
    echo "   - CPU maintained frequency targets."
fi

echo "------------------------------------------------------------"