RTNeural Comparisons

This repository contains an executable for comparing the performance of C++ neural network inferencing engines. Currently, there are four inferencing engines being compared:

• RTNeural (compile-time API)
• RTNeural (run-time API)
• libtorch
• onnxruntime
• TensorFlow Lite

Results

All benchmarks results were obtained on a 2018 Mac Mini with an Intel(R) Core(TM) i7-8700B CPU @ 3.20GHz. The "Real-Time Factor" measurement assumes a sample rate of 48 kHz.

Dense:

Conv1D:

GRU:

LSTM:

Activations:

Tanh:

ReLU:

Sigmoid: