API

NVML

 NVML ()

Convenient access to pynvml (the library behind nvidia-smi)

nv = NVML()
nv.driver_version, nv.cuda_version

('535.183.06', '12.2')

handle = nv[0]
nvmlDeviceGetName(handle)

'NVIDIA RTX A6000'

source

NVML.info

 NVML.info (i:int=0)

Basic information about GPU i

nv.info(0)

_Info(name='NVIDIA RTX A6000', serial='1322123048138', uuid='GPU-61e56e6f-2a64-c0f4-b26c-ab3ead0eed5b', persistence_mode=1, bus_id='00000000:01:00.0', display_active=0, performance_state=8, fan_speed=30, temperature=31, compute_mode=0)

source

NVML.mem

 NVML.mem (i:int=0)

Memory total/free/used for GPU i, in MB

nv.mem(0)

_Memory(free=2193.25, total=49140.0, used=46946.75)

source

NVML.utilization

 NVML.utilization (i:int=0)

% of time during which GPU i was actively using various components

nv.utilization(0)

_Utilization(gpu=0, memory=0, enc=0, dec=0)

source

NVML.power

 NVML.power (i:int=0)

Get power usage and limit for GPU i in watts

nv.power(0)

_Power(usage=16.866, limit=300.0)

Here “limit” refers to the maximum power draw allowed for the GPU.

source

NVML.clocks

 NVML.clocks (i:int=0)

Get current clock speeds (in MHz) for GPU i

nv.clocks(0)

_Clocks(graphics=0, sm=0, mem=405)

source

NVML.pcie_throughput

 NVML.pcie_throughput (i:int=0)

Get PCIe throughput (in KB/s) for GPU i

nv.pcie_throughput(0)

_PCIeThroughput(rx=0.0, tx=0.0)

“rx” represents receive - data flowing from the motherboard to the GPU.
“tx” represents transmit - data flowing from the GPU to the motherboard.

source

NVML.processes

 NVML.processes (i:int=0)

Get information about processes running on GPU i

nv.processes(0)

[_ProcessInfo(pid=201084, name='/home/jhoward/miniconda3/bin/python3.12', memory=46476.0)]

source

NVML.dmon

 NVML.dmon (i:int=0)

Get key monitoring metrics for GPU i, similar to nvidia-smi dmon

nv.dmon(0)

_DMon(pwr=16.866, gtemp=31, sm=0, mem=0, enc=0, dec=0, mclk=405, pclk=0)