def _update_nvml_static_info():
driver_version = pynvml.nvmlSystemGetDriverVersion().decode()
nvml_version = pynvml.nvmlSystemGetNVMLVersion().decode()
device_count = pynvml.nvmlDeviceGetCount()
devices = []
devices_handles = []
for i in range(device_count):
handle = pynvml.nvmlDeviceGetHandleByIndex(i)
name = pynvml.nvmlDeviceGetName(handle).decode()
mem_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
devices.append({
'index': i,
'name': name,
'memory': {
'total': mem_info.total
}
})
devices_handles.append(handle)
_static_info['public'].update({
'gpu': {
'driver': driver_version,
'nvml': nvml_version,
'devices': devices
}
})
_static_info['private'].update({'gpu': {'handles': devices_handles}})
def initialize(self, **kwargs) -> None:
try:
nvmlInit()
driver_version = nvmlSystemGetDriverVersion().decode("UTF-8")
nvml_version = nvmlSystemGetNVMLVersion().decode("UTF-8")
self.logger.info(f"NVML initialized, driver version: {driver_version}, NVML version: {nvml_version}")
self.detect_devices()
except NVMLError as error:
self.raise_nvml_error(error)
def get_nv_info():
nv_info = dict()
try:
nvmlInit()
nv_info["_Driver_Version"] = str(nvmlSystemGetDriverVersion(),
errors="ignore")
nv_info["_NVML_Version"] = str(nvmlSystemGetNVMLVersion(),
errors="ignore")
device_count = nvmlDeviceGetCount()
nv_info["Device_Count"] = device_count
devices = []
for i in range(device_count):
dev_info = dict()
handle = nvmlDeviceGetHandleByIndex(i)
dev_info["_Name"] = str(nvmlDeviceGetName(handle), errors="ignore")
memory_info = nvmlDeviceGetMemoryInfo(handle)
dev_info["Total_Memory"] = memory_info.total
dev_info["Free_Memory"] = memory_info.free
dev_info["Used_Memory"] = memory_info.used
util_rates = nvmlDeviceGetUtilizationRates(handle)
dev_info["GPU_Utilization_Rate"] = util_rates.gpu
dev_info["Memory_Utilization_Rate"] = util_rates.memory
devices.append(dev_info)
nv_info["Devices"] = devices
nvmlShutdown()
except Exception as e:
nv_info["Exception"] = str(e)
return nv_info
def test_nvmlSystemGetNVMLVersion(nvml):
vsn = 0.0
vsn = pynvml.nvmlSystemGetNVMLVersion().decode()
print("[NVML Version: " + vsn + "]", end=" ")
assert vsn > LooseVersion("0.0")
return {
'index': index,
'name': pynvml.nvmlDeviceGetName(handle).decode(),
'utilization': pynvml.nvmlDeviceGetUtilizationRates(handle).gpu,
'uuid': pynvml.nvmlDeviceGetUUID(handle).decode(),
}
if __name__ == '__main__':
pynvml.nvmlInit()
try:
print('NVIDIA Driver version %s' %
pynvml.nvmlSystemGetDriverVersion().decode())
print('NVML API %s initialized' %
pynvml.nvmlSystemGetNVMLVersion().decode())
print()
config = parse_arguments()
device_count = pynvml.nvmlDeviceGetCount()
devices = [query_device(i) for i in range(0, device_count)]
if not (config.device_index in range(0, device_count)):
print('Available devices:')
list(
map(
lambda device: print('Device #%d:\t%s\t%d%%\t%s' % (device[
'index'], device['name'], device[
'utilization'], device['uuid'])), devices))
exit(1)
def test_nvmlSystemGetNVMLVersion(nvml):
vsn = 0.0
vsn = pynvml.nvmlSystemGetNVMLVersion().decode()
print('[NVML Version: ' + vsn + ']', end=' ')
assert vsn > LooseVersion("0.0")