Executing Neural Networks on Low Powered GPUs

Honours Project Presentation

by Simon Rovder

Optimised Operations:

  • Matrix Multiplication
  • Bias Addition
  • Activation Functions (Sgimoid and ReLU)

Memory Layout Optimisation

Workgroup Size Optimisation

Blocking - Row Major by Row Major

Blocking - Row Major by Column Major

Executing Neural Networks on Low Powered GPUs

By simonrovder

Executing Neural Networks on Low Powered GPUs

  • 489