1580158
/
stable-diffusion-webui-GPU-temperature-protection
mirror of https://github.com/w-e-w/stable-diffusion-webui-GPU-temperature-protection.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

AMD & NVIDIA GPU OpenHardwareMonitor on Windows

pull/5/head
w-e-w 2023-09-15 19:49:15 +09:00
parent f7b1d5f17a
commit 012aeeaf43
6 changed files with 86 additions and 127 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.gitignore vendored
View File

@ -158,3 +158,5 @@ cython_debug/
# and can be added to the global gitignore or merged into this file. For a more nuclear
# option (not recommended) you can uncomment the following to ignore the entire idea folder.
#.idea/
OpenHardwareMonitor/OpenHardwareMonitorLib.dll

2
README.md
View File

@ -4,7 +4,7 @@ this extension uses nvidia-smi to monitor GPU temperature at the end of each ste
## Support
- Nvidia GPU on Windows and Linux
- AMD GPU on Linux with ROCm and Windows with openHardwareMonitor Lib
- AMD GPU on Linux with ROCm and Windows with [OpenHardwareMonitor](https://openhardwaremonitor.org/downloads)
## Installation
- method 1: Install using extensions tab `Available` extensions

19
scripts/constant.py
View File

@ -1,19 +0,0 @@
import os
from pathlib import Path
#OpenHardwareMonitorLib download url
OpenHardwareMonitorLibDownloadUrl = "https://openhardwaremonitor.org/files/openhardwaremonitor-v0.9.6.zip"
extensionRootPath = os.path.abspath(os.path.join(Path().absolute(),'./extensions/stable-diffusion-webui-GPU-temperature-protection/'))
#path of settings Storage
settingsStorageJsonPath = os.path.abspath(os.path.join(extensionRootPath,'./settingsStorage.json'))
# folder path
openHardwareMonitorDirPath = os.path.abspath(os.path.join(extensionRootPath,'./openHardwareMonitor'))
#path to load openHardwareMonitorLib using pythonnet
openHardwareMonitorLibPath = os.path.abspath(os.path.join(openHardwareMonitorDirPath,'./OpenHardwareMonitorLib'))
#dll path need for openhardwaremonitor zip extact
OpenHardwareMonitorLibdllFilePath = os.path.abspath(os.path.join(openHardwareMonitorDirPath,'./OpenHardwareMonitorLib.dll'))

126
scripts/gpu_temperature_protection.py
View File

@ -1,19 +1,28 @@
from modules import scripts, shared, sd_samplers_common
from pathlib import Path
import urllib.request
import gradio as gr
import subprocess
import zipfile
import launch
import time
import re
import os
from scripts import constant,settings_storage,install
import launch
if shared.opts.gpu_temps_sleep_temperature_src == 'NVIDIA & AMD - openHardwareMonitor' and os.name == 'nt' and launch.is_installed("pythonnet"):
# check is OpenHardwareMonitorLib exist if not will download
if not os.path.isfile(constant.OpenHardwareMonitorLibdllFilePath):
install.downloadOpenHardwareMonitorLib()
import clr # the pythonnet module.
clr.AddReference(constant.openHardwareMonitorLibPath)
from OpenHardwareMonitor.Hardware import Computer
OpenHardwareMonitorLibDownloadUrl = "https://openhardwaremonitor.org/files/openhardwaremonitor-v0.9.6.zip"
OpenHardwareMonitor_path = Path(scripts.current_basedir).joinpath('OpenHardwareMonitor')
OpenHardwareMonitorLib_path = OpenHardwareMonitor_path.joinpath('OpenHardwareMonitorLib')
OpenHardwareMonitorLib_dll_path = OpenHardwareMonitor_path.joinpath('OpenHardwareMonitorLib.dll')
def download_open_hardware_monitor():
if not OpenHardwareMonitorLib_dll_path.is_file():
OpenHardwareMonitor_path.mkdir(parents=True, exist_ok=True)
print("Downloading OpenHardwareMonitor")
zip_path, _ = urllib.request.urlretrieve(OpenHardwareMonitorLibDownloadUrl)
with zipfile.ZipFile(zip_path, "r") as z:
with open(os.path.realpath(OpenHardwareMonitorLib_dll_path), 'wb') as f:
f.write(z.read('OpenHardwareMonitor/OpenHardwareMonitorLib.dll'))
class GPUTemperatureProtection(scripts.Script):
@ -64,44 +73,66 @@ class GPUTemperatureProtection(scripts.Script):
hardware = None
@staticmethod
def get_gpu_temperature_open_hardware_monitor():
if not launch.is_installed("pythonnet"):
print("\n[Error GPU temperature protection] openHardwareMonitor : you need restart to install and download requirement")
if os.name != "nt":
print("\n[Error GPU temperature protection] openHardwareMonitor : only works on windows")
return 0
def init_open_hardware_monitor():
try:
if GPUTemperatureProtection.computer is None:
GPUTemperatureProtection.computer = Computer()
GPUTemperatureProtection.computer.CPUEnabled = True # get the Info about CPU
GPUTemperatureProtection.computer.GPUEnabled = True # get the Info about GPU
GPUTemperatureProtection.computer.Open()
if GPUTemperatureProtection.sensors is None:
for a in range(0, len(GPUTemperatureProtection.computer.Hardware)):
if shared.opts.gpu_temps_sleep_gpu_name in str(GPUTemperatureProtection.computer.Hardware[a].Name):
for b in range(0, len(GPUTemperatureProtection.computer.Hardware[a].Sensors)):
if "/temperature" in str(GPUTemperatureProtection.computer.Hardware[a].Sensors[b].Identifier):
GPUTemperatureProtection.sensors = GPUTemperatureProtection.computer.Hardware[a].Sensors[b]
GPUTemperatureProtection.hardware = GPUTemperatureProtection.computer.Hardware[a]
if GPUTemperatureProtection.sensors is None:
print("\n[Error GPU temperature protection] openHardwareMonitor : Couldn't read temperature from OpenHardwareMonitorLib")
return 0
# install and import Python.NET module
if not launch.is_installed("pythonnet"):
launch.run_pip("install pythonnet==3.0.2", "Installing requirements for OpenHardwareMonitorLib")
import clr # import pythonnet module.
# download OpenHardwareMonitor if not found
download_open_hardware_monitor()
# initialize OpenHardwareMonitor
if GPUTemperatureProtection.computer is None:
clr.AddReference(str(OpenHardwareMonitorLib_path))
from OpenHardwareMonitor.Hardware import Computer
GPUTemperatureProtection.computer = Computer()
GPUTemperatureProtection.computer.CPUEnabled = False # Disable CPU
GPUTemperatureProtection.computer.GPUEnabled = True # Enable GPU
GPUTemperatureProtection.computer.Open()
# find the first matching temperature sensor for the specified hardware
if GPUTemperatureProtection.sensors is None or shared.opts.gpu_temps_sleep_gpu_name not in str(GPUTemperatureProtection.hardware.Name):
for hardware in GPUTemperatureProtection.computer.Hardware:
if shared.opts.gpu_temps_sleep_gpu_name in str(hardware.Name):
for sensor in hardware.Sensors:
if '/temperature' in str(sensor.Identifier):
GPUTemperatureProtection.sensors = sensor
GPUTemperatureProtection.hardware = hardware
return # sensor is found early return
# sensor not found
GPUTemperatureProtection.sensors = None
GPUTemperatureProtection.hardware = None
print(f"[Error GPU temperature protection] OpenHardwareMonitor Couldn't find temperature sensor for {shared.opts.gpu_temps_sleep_gpu_name}")
except Exception as e:
print(f"[Error GPU temperature protection] Failed to initialize OpenHardwareMonitor \: {e}")
@staticmethod
def get_gpu_temperature_open_hardware_monitor():
try:
GPUTemperatureProtection.hardware.Update()
return int(GPUTemperatureProtection.sensors.get_Value())
except Exception as e:
print(f'\n[Error GPU temperature protection] openHardwareMonitor : {e}')
print(f"\n[Error GPU temperature protection] OpenHardwareMonitor: Couldn't read temperature{e}")
return 0
@staticmethod
def onChangeGpuTempsSleepTemperatureSrc():
settings_storage.settingsStorage.set("gpu_temps_sleep_temperature_src",shared.opts.gpu_temps_sleep_temperature_src)
settings_storage.settingsStorage.save()
def on_change_temps_src():
if shared.opts.gpu_temps_sleep_temperature_src == 'NVIDIA & AMD - OpenHardwareMonitor':
if os.name == 'nt':
GPUTemperatureProtection.init_open_hardware_monitor()
else:
assert False, "NVIDIA & AMD - OpenHardwareMonitor it's only supported on Windows"
elif shared.opts.gpu_temps_sleep_temperature_src == 'AMD - ROCm-smi' and os.name == 'nt':
assert False, "AMD - ROCm-smi is not supported on Windows"
temperature_src_dict = {
"NVIDIA - nvidia-smi": get_gpu_temperature_nvidia_smi,
"AMD - ROCm-smi": get_gpu_temperature_amd_rocm_smi,
"NVIDIA & AMD - openHardwareMonitor": get_gpu_temperature_open_hardware_monitor
"NVIDIA & AMD - OpenHardwareMonitor": get_gpu_temperature_open_hardware_monitor
}
@staticmethod
@ -156,12 +187,13 @@ if hasattr(shared, "OptionHTML"): # < 1.6.0 support
shared.options_templates.update(shared.options_section(('GPU_temperature_protection', "GPU Temperature"), {
"gpu_temps_sleep_temperature_src_explanation": shared.OptionHTML("""<b>NVIDIA - nvidia-smi</b> is available on both Windows and Linux.<br>
<b>AMD - ROCm-smi</b> is Linux only and does not support specifying GPU device index.<br>
<b>NVIDIA & AMD - openHardwareMonitor</b> is windows only suport NVIDIA and AMD.
<b>NVIDIA & AMD - OpenHardwareMonitor</b> is Windows only supports NVIDIA and AMD.
""")
}))
shared.options_templates.update(shared.options_section(('GPU_temperature_protection', "GPU Temperature"), {
"gpu_temps_sleep_temperature_src": shared.OptionInfo("NVIDIA - nvidia-smi", "Temperature source", gr.Radio, {"choices": list(GPUTemperatureProtection.temperature_src_dict.keys())}, onchange=GPUTemperatureProtection.onChangeGpuTempsSleepTemperatureSrc).needs_restart(),
"gpu_temps_sleep_temperature_src": shared.OptionInfo("NVIDIA - nvidia-smi", "Temperature source", gr.Radio, {"choices": list(GPUTemperatureProtection.temperature_src_dict.keys())}, GPUTemperatureProtection.on_change_temps_src),
"gpu_temps_sleep_enable": shared.OptionInfo(True, "Enable GPU temperature protection"),
"gpu_temps_sleep_print": shared.OptionInfo(True, "Print GPU Core temperature while sleeping in terminal"),
"gpu_temps_sleep_minimum_interval": shared.OptionInfo(5.0, "GPU temperature monitor minimum interval", gr.Number).info("won't check the temperature again until this amount of seconds have passed"),
@ -169,12 +201,20 @@ shared.options_templates.update(shared.options_section(('GPU_temperature_protect
"gpu_temps_sleep_max_sleep_time": shared.OptionInfo(10.0, "Max sleep Time", gr.Number).info("max number of seconds that it's allowed to pause, 0=unlimited"),
"gpu_temps_sleep_sleep_temp": shared.OptionInfo(75.0, "GPU sleep temperature", gr.Slider, {"minimum": 0, "maximum": 125}).info("generation will pause if GPU core temperature exceeds this temperature"),
"gpu_temps_sleep_wake_temp": shared.OptionInfo(75.0, "GPU wake temperature", gr.Slider, {"minimum": 0, "maximum": 125}).info("generation will pause until GPU core temperature drops below this temperature"),
"gpu_temps_sleep_gpu_index": shared.OptionInfo(0, "GPU device index", gr.Number, {"precision": 0}).info("selecting the correct temperature reading for multi GPU systems, for systems with 3 gpus the value should be an integer between 0~2, default 0"),
"gpu_temps_sleep_gpu_index": shared.OptionInfo(0, "GPU device index - nvidia-smi", gr.Number, {"precision": 0}).info("selecting the correct temperature reading for multi GPU systems, for systems with 3 gpus the value should be an integer between 0~2, default 0"),
}))
if os.name == 'nt':
all_lines = subprocess.check_output(['cmd.exe', '/c', 'wmic path win32_VideoController get name']).decode().strip("\nName").splitlines()
names_list = [name.rstrip() for name in all_lines if not re.compile("^ +$").match(name) and name != '']
shared.options_templates.update(shared.options_section(('GPU_temperature_protection', "GPU Temperature"), {
"gpu_temps_sleep_gpu_name": shared.OptionInfo( "none" if len(names_list) == 0 else names_list[0] , "GPU Name", gr.Radio, {"choices": names_list, "interactive":shared.opts.gpu_temps_sleep_temperature_src == 'NVIDIA & AMD - openHardwareMonitor' }).info("select your gpu, <b>only for openHardwareMonitor - windows</b>"),
}))
try:
all_lines = subprocess.check_output(['cmd.exe', '/c', 'wmic path win32_VideoController get name']).decode().strip("\nName").splitlines()
video_controller_filter = re.compile(r"^\s+$")
names_list = [name.strip() for name in all_lines if not video_controller_filter.match(name) and name != '']
shared.options_templates.update(shared.options_section(('GPU_temperature_protection', "GPU Temperature"), {
"gpu_temps_sleep_gpu_name": shared.OptionInfo("None" if len(names_list) == 0 else names_list[0], "GPU Name - OpenHardwareMonitor", gr.Radio, {"choices": names_list}, GPUTemperatureProtection.on_change_temps_src).info("select your gpu"),
}))
except Exception as _e:
if shared.opts.gpu_temps_sleep_temperature_src == 'NVIDIA & AMD - OpenHardwareMonitor':
print(f'[Error GPU temperature protection] Failed to retrieve list of video controllers: \n{_e}')
if shared.opts.gpu_temps_sleep_temperature_src == 'NVIDIA & AMD - OpenHardwareMonitor':
GPUTemperatureProtection.init_open_hardware_monitor()

26
scripts/install.py
View File

@ -1,26 +0,0 @@
import launch
import os
from pathlib import Path
from scripts import constant,settings_storage
import urllib
import zipfile
def downloadOpenHardwareMonitorLib():
zip_path, _ = urllib.request.urlretrieve(constant.OpenHardwareMonitorLibDownloadUrl)
with zipfile.ZipFile(zip_path, "r") as z:
with open(os.path.realpath(constant.OpenHardwareMonitorLibdllFilePath) , 'wb') as f:
f.write(z.read('OpenHardwareMonitor/OpenHardwareMonitorLib.dll'))
# install pythonnet required for openHardwareMonitor Lib
# shared.opts.gpu_temps_sleep_temperature_src == 'NVIDIA & AMD - openHardwareMonitor' and
if os.name == 'nt' :
if settings_storage.settingsStorage.get("gpu_temps_sleep_temperature_src") == "NVIDIA & AMD - openHardwareMonitor":
if not launch.is_installed("pythonnet"):
launch.run_pip("install pythonnet==3.0.2", "requirements for windows OpenHardwareMonitorLib")
# check and create OpenHardwareMonitor folder
Path(constant.openHardwareMonitorDirPath).mkdir(parents=True, exist_ok=True)

38
scripts/settings_storage.py
View File

@ -1,38 +0,0 @@
import json
from scripts import constant
import os
class _SettingsStorage:
def __init__(self,):
self.json_decoded = json.loads("{}")
# check if not exist create empty file
if not os.path.isfile(constant.settingsStorageJsonPath) or not os.access(constant.settingsStorageJsonPath, os.R_OK):
with open(constant.settingsStorageJsonPath, 'w') as json_file:
json_file.write(json.dumps({}))
json_file.close()
with open(constant.settingsStorageJsonPath) as json_file:
self.json_decoded = json.loads(json_file.read())
json_file.close()
def save(self):
with open(constant.settingsStorageJsonPath, 'w') as json_file:
json_file.write(json.dumps(self.json_decoded))
#json.dump(self.json_decoded, json_file)
json_file.close()
def get(self, key):
if key in self.json_decoded:
return self.json_decoded[key]
return None
def set(self, key, val):
self.json_decoded[key] = val
settingsStorage = _SettingsStorage()