maxDNN: An Efficient Convolution Kernel for Deep Learning with Maxwell GPUs
This work provides a domain-specific optimization for deep learning practitioners using Maxwell GPUs, but it is incremental as it builds on existing methods like cuda-convnet2 and Maxas SGEMM.
The paper tackles the problem of computationally efficient convolution for deep learning on NVIDIA Maxwell GPUs by introducing maxDNN, which achieves 96.3% computational efficiency on typical network architectures.
This paper describes maxDNN, a computationally efficient convolution kernel for deep learning with the NVIDIA Maxwell GPU. maxDNN reaches 96.3% computational efficiency on typical deep learning network architectures. The design combines ideas from cuda-convnet2 with the Maxas SGEMM assembly code. We only address forward propagation (FPROP) operation of the network, but we believe that the same techniques used here will be effective for backward propagation (BPROP) as well.