Example showing how to utilize NVML library for GPU monitoring with high sampling rate.

This example utilizing the NVML Library and C++11 mutlithreading to provide GPU monitoring with a high sampling rate. Normally, one would pipe nvidia-smi to a file, but this can cause excessive I/O usage.

Also provided is a Matlab script used to plot the data.

Example requires Tesla card because some calls are not supported on Geforce family.

This example utilizes the following toolsets:

• cuBLAS
• Thrust
• C++11 multithreading
• NVIDIA Management Library (NVML)

Include header class

#include nvmlClass.h

Create object of nvmlClass

nvmlClass nvml( dev, filename );

Create thread to run in parallel, gathering data

std::thread threadStart( &nvmlClass::getStats, &nvml );

Create thread to kill getStats function

std::thread threadKill( &nvmlClass::killThread, &nvml );

Join all threads

threadStart.join( );

threadKill.join( );