|
| 1 | +#!/bin/bash |
| 2 | + |
| 3 | +# 默认参数 |
| 4 | +device_id=0 |
| 5 | +interval=1 |
| 6 | +count=-1 |
| 7 | +output_file="gpu_monitor_log.txt" |
| 8 | + |
| 9 | +# ANSI 颜色 |
| 10 | +RED='\033[0;31m' |
| 11 | +NC='\033[0m' # 清除颜色 |
| 12 | + |
| 13 | +# 报警计数器 |
| 14 | +high_temp_count=0 |
| 15 | +low_freq_count=0 |
| 16 | + |
| 17 | +# 解析命令行参数(不包括 -o) |
| 18 | +while getopts "d:i:n:" opt; do |
| 19 | + case $opt in |
| 20 | + d) device_id=$OPTARG ;; |
| 21 | + i) interval=$OPTARG ;; |
| 22 | + n) count=$OPTARG ;; |
| 23 | + *) echo "Usage: $0 [-d device_id] [-i interval_sec] [-n count]" |
| 24 | + exit 1 ;; |
| 25 | + esac |
| 26 | +done |
| 27 | + |
| 28 | +# 写入启动信息(仅首次) |
| 29 | +if [ ! -f "$output_file" ]; then |
| 30 | + echo "=== GPU Monitor Started at $(date) ===" >> "$output_file" |
| 31 | + echo "Device: $device_id, Interval: $interval sec" >> "$output_file" |
| 32 | + echo "----------------------------------------" >> "$output_file" |
| 33 | +fi |
| 34 | + |
| 35 | +# 开始监控 |
| 36 | +i=0 |
| 37 | +while [ $count -lt 0 ] || [ $i -lt $count ]; do |
| 38 | + timestamp=$(date "+%Y-%m-%d %H:%M:%S") |
| 39 | + temp_line=$(mthreads-gmi -i "$device_id" -q | grep "GPU Current Temp") |
| 40 | + freq_line=$(mthreads-gmi -i "$device_id" -q | grep "Graphics") |
| 41 | + |
| 42 | + temp=$(echo "$temp_line" | grep -oP '\d+(?=C)') |
| 43 | + freq=$(echo "$freq_line" | grep -oP '\d+(?=MHz)') |
| 44 | + |
| 45 | + plain_msg="" |
| 46 | + color_msg="" |
| 47 | + |
| 48 | + # 检查高温 |
| 49 | + if [ "$temp" -gt 95 ]; then |
| 50 | + plain_msg+="高温报警(${temp}C)" |
| 51 | + color_msg+="${RED}高温报警(${temp}C)${NC} " |
| 52 | + ((high_temp_count++)) |
| 53 | + fi |
| 54 | + |
| 55 | + # 检查降频 |
| 56 | + if [ "$freq" -lt 1750 ]; then |
| 57 | + plain_msg+=" 降频警告(${freq}MHz)" |
| 58 | + color_msg+="${RED}降频警告(${freq}MHz)${NC}" |
| 59 | + ((low_freq_count++)) |
| 60 | + fi |
| 61 | + |
| 62 | + { |
| 63 | + echo "[$timestamp]" |
| 64 | + echo "$temp_line" |
| 65 | + echo "$freq_line" |
| 66 | + [ -n "$plain_msg" ] && echo "$plain_msg" |
| 67 | + echo "------------------------------------" |
| 68 | + } >> "$output_file" |
| 69 | + |
| 70 | + # 终端打印报警(红色) |
| 71 | + [ -n "$plain_msg" ] && echo -e "[$timestamp] $color_msg" |
| 72 | + |
| 73 | + sleep "$interval" |
| 74 | + ((i++)) |
| 75 | +done |
| 76 | + |
| 77 | +# 打印统计结果 |
| 78 | +echo -e "\n=== 监控结束 ===" |
| 79 | +echo "高温次数: $high_temp_count" |
| 80 | +echo "降频次数: $low_freq_count" |
| 81 | + |
0 commit comments