Fast Convolutional Nets With fbfft: A GPU Performance Evaluation
This work provides faster convolution implementations for deep learning practitioners using NVIDIA GPUs, though it is incremental as it builds on existing FFT methods.
The authors evaluated GPU performance for convolutional neural network training and introduced two FFT-based convolution implementations (cuFFT and fbfft) that achieve speedups of over 1.5x over cuFFT for whole CNNs and up to 23.5x over cuDNN for some synthetic kernel configurations.
We examine the performance profile of Convolutional Neural Network training on the current generation of NVIDIA Graphics Processing Units. We introduce two new Fast Fourier Transform convolution implementations: one based on NVIDIA's cuFFT library, and another based on a Facebook authored FFT implementation, fbfft, that provides significant speedups over cuFFT (over 1.5x) for whole CNNs. Both of these convolution implementations are available in open source, and are faster than NVIDIA's cuDNN implementation for many common convolutional layers (up to 23.5x for some synthetic kernel configurations). We discuss different performance regimes of convolutions, comparing areas where straightforward time domain convolutions outperform Fourier frequency domain convolutions. Details on algorithmic applications of NVIDIA GPU hardware specifics in the implementation of fbfft are also provided.