def brand(self):
""" Device brand name as a string
This function maps the device code to a string representation using the
following enum:
NVML_BRAND_UNKNOWN = 0
NVML_BRAND_QUADRO = 1
NVML_BRAND_TESLA = 2
NVML_BRAND_NVS = 3
NVML_BRAND_GRID = 4
NVML_BRAND_GEFORCE = 5
NVML_BRAND_TITAN = 6
"""
brand_enum = nv.nvmlDeviceGetBrand(self._handle)
if brand_enum == 1:
return 'Quadro'
elif brand_enum == 2:
return 'Tesla'
elif brand_enum == 3:
return 'NVS'
elif brand_enum == 4:
return 'Grid'
elif brand_enum == 5:
return 'GeForce'
elif brand_enum == 6:
return 'Titan'
else:
return 'Unknown'
def _get_data(self):
data = {}
if self.deviceCount:
for i in range(self.deviceCount):
gpuIdx = str(i)
handle = pynvml.nvmlDeviceGetHandleByIndex(i)
name = pynvml.nvmlDeviceGetName(handle)
brand = pynvml.nvmlDeviceGetBrand(handle)
### Get data ###
## Memory usage
try:
mem = pynvml.nvmlDeviceGetMemoryInfo(handle)
except Exception as e:
self.debug(str(e))
mem = None
## ECC errors
try:
_memError = {}
_eccCounter = {}
eccErrors = {}
eccCounterType = ['VOLATILE_ECC', 'AGGREGATE_ECC']
memErrorType = [
'ERROR_TYPE_CORRECTED', 'ERROR_TYPE_UNCORRECTED'
]
memoryLocationType = [
'L1_CACHE', 'L2_CACHE', 'DEVICE_MEMORY',
'REGISTER_FILE', 'TEXTURE_MEMORY'
]
for memoryLocation in range(5):
for eccCounter in range(2):
for memError in range(2):
_memError[memErrorType[
memError]] = pynvml.nvmlDeviceGetMemoryErrorCounter(
handle, memError, eccCounter,
memoryLocation)
_eccCounter[eccCounterType[eccCounter]] = _memError
eccErrors[
memoryLocationType[memoryLocation]] = _eccCounter
except Exception as e:
self.debug(str(e))
eccErrors = None
## Temperature
try:
temp = pynvml.nvmlDeviceGetTemperature(
handle, pynvml.NVML_TEMPERATURE_GPU)
except Exception as e:
self.debug(str(e))
temp = None
## Fan
try:
fanspeed = pynvml.nvmlDeviceGetFanSpeed(handle)
except Exception as e:
self.debug(str(e))
fanspeed = None
## GPU and Memory Utilization
try:
util = pynvml.nvmlDeviceGetUtilizationRates(handle)
gpu_util = util.gpu
mem_util = util.memory
except Exception as e:
self.debug(str(e))
gpu_util = None
mem_util = None
## Encoder Utilization
try:
encoder = pynvml.nvmlDeviceGetEncoderUtilization(handle)
enc_util = encoder[0]
except Exception as e:
self.debug(str(e))
enc_util = None
## Decoder Utilization
try:
decoder = pynvml.nvmlDeviceGetDecoderUtilization(handle)
dec_util = decoder[0]
except Exception as e:
self.debug(str(e))
dec_util = None
## Clock frequencies
try:
clock_core = pynvml.nvmlDeviceGetClockInfo(
handle, pynvml.NVML_CLOCK_GRAPHICS)
clock_sm = pynvml.nvmlDeviceGetClockInfo(
handle, pynvml.NVML_CLOCK_SM)
clock_mem = pynvml.nvmlDeviceGetClockInfo(
handle, pynvml.NVML_CLOCK_MEM) * self.nvMemFactor
except Exception as e:
self.debug(str(e))
clock_core = None
clock_sm = None
clock_mem = None
### Packing data ###
self.debug("Device", gpuIdx, ":", str(name))
data["device_name_" + gpuIdx] = name
self.debug("Brand:", str(brand))
self.debug(str(name), "Temp :", str(temp))
data["device_temp_" + gpuIdx] = temp
self.debug(str(name), "Mem total :", str(mem.total), 'bytes')
data["device_mem_total_" + gpuIdx] = mem.total
self.debug(str(name), "Mem used :", str(mem.used), 'bytes')
data["device_mem_used_" + gpuIdx] = mem.used
self.debug(str(name), "Mem free :", str(mem.free), 'bytes')
data["device_mem_free_" + gpuIdx] = mem.free
self.debug(str(name), "Load GPU :", str(gpu_util), '%')
data["device_load_gpu_" + gpuIdx] = gpu_util
self.debug(str(name), "Load MEM :", str(mem_util), '%')
data["device_load_mem_" + gpuIdx] = mem_util
self.debug(str(name), "Load ENC :", str(enc_util), '%')
data["device_load_enc_" + gpuIdx] = enc_util
self.debug(str(name), "Load DEC :", str(dec_util), '%')
data["device_load_dec_" + gpuIdx] = dec_util
self.debug(str(name), "Core clock:", str(clock_core), 'MHz')
data["device_core_clock_" + gpuIdx] = clock_core
self.debug(str(name), "SM clock :", str(clock_sm), 'MHz')
data["device_sm_clock_" + gpuIdx] = clock_sm
self.debug(str(name), "Mem clock :", str(clock_mem), 'MHz')
data["device_mem_clock_" + gpuIdx] = clock_mem
self.debug(str(name), "Fan speed :", str(fanspeed), '%')
data["device_fanspeed_" + gpuIdx] = fanspeed
self.debug(str(name), "ECC errors:", str(eccErrors))
if eccErrors is not None:
data["device_ecc_errors_L1_CACHE_VOLATILE_CORRECTED_" +
gpuIdx] = eccErrors["L1_CACHE"]["VOLATILE_ECC"][
"ERROR_TYPE_CORRECTED"]
data["device_ecc_errors_L1_CACHE_VOLATILE_UNCORRECTED_" +
gpuIdx] = eccErrors["L1_CACHE"]["VOLATILE_ECC"][
"ERROR_TYPE_UNCORRECTED"]
data["device_ecc_errors_L1_CACHE_AGGREGATE_CORRECTED_" +
gpuIdx] = eccErrors["L1_CACHE"]["AGGREGATE_ECC"][
"ERROR_TYPE_CORRECTED"]
data["device_ecc_errors_L1_CACHE_AGGREGATE_UNCORRECTED_" +
gpuIdx] = eccErrors["L1_CACHE"]["AGGREGATE_ECC"][
"ERROR_TYPE_UNCORRECTED"]
data["device_ecc_errors_L2_CACHE_VOLATILE_CORRECTED_" +
gpuIdx] = eccErrors["L2_CACHE"]["VOLATILE_ECC"][
"ERROR_TYPE_CORRECTED"]
data["device_ecc_errors_L2_CACHE_VOLATILE_UNCORRECTED_" +
gpuIdx] = eccErrors["L2_CACHE"]["VOLATILE_ECC"][
"ERROR_TYPE_UNCORRECTED"]
data["device_ecc_errors_L2_CACHE_AGGREGATE_CORRECTED_" +
gpuIdx] = eccErrors["L2_CACHE"]["AGGREGATE_ECC"][
"ERROR_TYPE_CORRECTED"]
data["device_ecc_errors_L2_CACHE_AGGREGATE_UNCORRECTED_" +
gpuIdx] = eccErrors["L2_CACHE"]["AGGREGATE_ECC"][
"ERROR_TYPE_UNCORRECTED"]
data["device_ecc_errors_DEVICE_MEMORY_VOLATILE_CORRECTED_"
+ gpuIdx] = eccErrors["DEVICE_MEMORY"][
"VOLATILE_ECC"]["ERROR_TYPE_CORRECTED"]
data[
"device_ecc_errors_DEVICE_MEMORY_VOLATILE_UNCORRECTED_"
+ gpuIdx] = eccErrors["DEVICE_MEMORY"]["VOLATILE_ECC"][
"ERROR_TYPE_UNCORRECTED"]
data["device_ecc_errors_DEVICE_MEMORY_AGGREGATE_CORRECTED_"
+ gpuIdx] = eccErrors["DEVICE_MEMORY"][
"AGGREGATE_ECC"]["ERROR_TYPE_CORRECTED"]
data[
"device_ecc_errors_DEVICE_MEMORY_AGGREGATE_UNCORRECTED_"
+ gpuIdx] = eccErrors["DEVICE_MEMORY"][
"AGGREGATE_ECC"]["ERROR_TYPE_UNCORRECTED"]
data["device_ecc_errors_REGISTER_FILE_VOLATILE_CORRECTED_"
+ gpuIdx] = eccErrors["REGISTER_FILE"][
"VOLATILE_ECC"]["ERROR_TYPE_CORRECTED"]
data[
"device_ecc_errors_REGISTER_FILE_VOLATILE_UNCORRECTED_"
+ gpuIdx] = eccErrors["REGISTER_FILE"]["VOLATILE_ECC"][
"ERROR_TYPE_UNCORRECTED"]
data["device_ecc_errors_REGISTER_FILE_AGGREGATE_CORRECTED_"
+ gpuIdx] = eccErrors["REGISTER_FILE"][
"AGGREGATE_ECC"]["ERROR_TYPE_CORRECTED"]
data[
"device_ecc_errors_REGISTER_FILE_AGGREGATE_UNCORRECTED_"
+ gpuIdx] = eccErrors["REGISTER_FILE"][
"AGGREGATE_ECC"]["ERROR_TYPE_UNCORRECTED"]
data["device_ecc_errors_TEXTURE_MEMORY_VOLATILE_CORRECTED_"
+ gpuIdx] = eccErrors["TEXTURE_MEMORY"][
"VOLATILE_ECC"]["ERROR_TYPE_CORRECTED"]
data[
"device_ecc_errors_TEXTURE_MEMORY_VOLATILE_UNCORRECTED_"
+ gpuIdx] = eccErrors["TEXTURE_MEMORY"][
"VOLATILE_ECC"]["ERROR_TYPE_UNCORRECTED"]
data[
"device_ecc_errors_TEXTURE_MEMORY_AGGREGATE_CORRECTED_"
+ gpuIdx] = eccErrors["TEXTURE_MEMORY"][
"AGGREGATE_ECC"]["ERROR_TYPE_CORRECTED"]
data[
"device_ecc_errors_TEXTURE_MEMORY_AGGREGATE_UNCORRECTED_"
+ gpuIdx] = eccErrors["TEXTURE_MEMORY"][
"AGGREGATE_ECC"]["ERROR_TYPE_UNCORRECTED"]
else:
data["device_ecc_errors_L1_CACHE_VOLATILE_CORRECTED_" +
gpuIdx] = None
## Get unit (S-class Nvidia cards) data
if self.unitCount:
for i in range(self.unitCount):
gpuIdx = str(i)
handle = pynvml.nvmlUnitGetHandleByIndex(i)
try:
fan = pynvml.nvmlUnitGetFanSpeedInfo(handle)
fan_speed = fan.speed # Fan speed (RPM)
fan_state = fan.state # Flag that indicates whether fan is working properly
except Exception as e:
self.debug(str(e))
fan_speed = None
fan_state = None
try:
psu = pynvml.nvmlUnitGetPsuInfo(handle)
psu_current = psu.current # PSU current (A)
psu_power = psu.power # PSU power draw (W)
psu_state = psu.state # The power supply state
psu_voltage = psu.voltage # PSU voltage (V)
except Exception as e:
self.debug(str(e))
psu_current = None
psu_power = None
psu_state = None
psu_voltage = None
try:
temp_intake = pynvml.nvmlUnitGetTemperature(
handle, 0) # Temperature at intake in C
temp_exhaust = pynvml.nvmlUnitGetTemperature(
handle, 1) # Temperature at exhaust in C
temp_board = pynvml.nvmlUnitGetTemperature(
handle, 2) # Temperature on board in C
except Exception as e:
self.debug(str(e))
temp_intake = None
temp_exhaust = None
temp_board = None
self.debug('Unit fan speed:', str(fan_speed))
data["unit_fan_speed_" + gpuIdx] = fan_speed
self.debug('Unit fan state:', str(fan_state))
data["unit_fan_state_" + gpuIdx] = fan_state
self.debug('Unit PSU current:', str(psu_current))
data["unit_psu_current_" + gpuIdx] = psu_current
self.debug('Unit PSU power:', str(psu_power))
data["unit_psu_power_" + gpuIdx] = psu_power
self.debug('Unit PSU state:', str(psu_state))
data["unit_psu_state_" + gpuIdx] = psu_state
self.debug('Unit PSU voltage:', str(psu_voltage))
data["unit_psu_voltage_" + gpuIdx] = psu_voltage
self.debug('Unit temp intake:', str(temp_intake))
data["unit_temp_intake_" + gpuIdx] = temp_intake
self.debug('Unit temp exhaust:', str(temp_exhaust))
data["unit_temp_exhaust_" + gpuIdx] = temp_exhaust
self.debug('Unit temp board:', str(temp_board))
data["unit_temp_board_" + gpuIdx] = temp_board
## Get data via legacy mode
if self.legacy:
try:
output, error = Popen([
"nvidia-settings", "-c", ":0", "-q", "GPUUtilization",
"-q", "GPUCurrentClockFreqs", "-q", "GPUCoreTemp", "-q",
"TotalDedicatedGPUMemory", "-q", "UsedDedicatedGPUMemory"
],
shell=False,
stdout=PIPE,
stderr=PIPE).communicate()
output = repr(str(output))
if len(output) < 800:
raise Exception(
'Error in fetching data from nvidia-settings ' +
output)
self.debug(str(error), output)
except Exception as e:
self.error(str(e))
self.error('Setting legacy mode to False')
self.legacy = False
return data
for i in range(self.deviceCount):
gpuIdx = str(i)
if data["device_temp_" + gpuIdx] is None:
coreTemp = findall('GPUCoreTemp.*?(gpu:\d*).*?\s(\d*)',
output)[i][1]
try:
data["device_temp_" + gpuIdx] = int(coreTemp)
self.debug('Using legacy temp for GPU {0}: {1}'.format(
gpuIdx, coreTemp))
except Exception as e:
self.debug(str(e), "skipping device_temp_" + gpuIdx)
if data["device_mem_used_" + gpuIdx] is None:
memUsed = findall(
'UsedDedicatedGPUMemory.*?(gpu:\d*).*?\s(\d*)',
output)[i][1]
try:
data["device_mem_used_" + gpuIdx] = int(memUsed)
self.debug(
'Using legacy mem_used for GPU {0}: {1}'.format(
gpuIdx, memUsed))
except Exception as e:
self.debug(str(e),
"skipping device_mem_used_" + gpuIdx)
if data["device_load_gpu_" + gpuIdx] is None:
gpu_util = findall(
'(gpu:\d*).*?graphics=(\d*),.*?memory=(\d*)',
output)[i][1]
try:
data["device_load_gpu_" + gpuIdx] = int(gpu_util)
self.debug(
'Using legacy load_gpu for GPU {0}: {1}'.format(
gpuIdx, gpu_util))
except Exception as e:
self.debug(str(e),
"skipping device_load_gpu_" + gpuIdx)
if data["device_load_mem_" + gpuIdx] is None:
mem_util = findall(
'(gpu:\d*).*?graphics=(\d*),.*?memory=(\d*)',
output)[i][2]
try:
data["device_load_mem_" + gpuIdx] = int(mem_util)
self.debug(
'Using legacy load_mem for GPU {0}: {1}'.format(
gpuIdx, mem_util))
except Exception as e:
self.debug(str(e),
"skipping device_load_mem_" + gpuIdx)
if data["device_core_clock_" + gpuIdx] is None:
clock_core = findall(
'GPUCurrentClockFreqs.*?(gpu:\d*).*?(\d*),(\d*)',
output)[i][1]
try:
data["device_core_clock_" + gpuIdx] = int(clock_core)
self.debug(
'Using legacy core_clock for GPU {0}: {1}'.format(
gpuIdx, clock_core))
except Exception as e:
self.debug(str(e),
"skipping device_core_clock_" + gpuIdx)
if data["device_mem_clock_" + gpuIdx] is None:
clock_mem = findall(
'GPUCurrentClockFreqs.*?(gpu:\d*).*?(\d*),(\d*)',
output)[i][2]
try:
data["device_mem_clock_" + gpuIdx] = int(clock_mem)
self.debug(
'Using legacy mem_clock for GPU {0}: {1}'.format(
gpuIdx, clock_mem))
except Exception as e:
self.debug(str(e),
"skipping device_mem_clock_" + gpuIdx)
return data