cuDNN: Efficient Primitives for Deep Learning

We present a library that provides optimized implementations for deep learning primitives. Deep learning workloads are computationally intensive, and optimiz-ing the kernels of deep learning workloads is difficult and time-consuming. As parallel architectures evolve, kernels must be reoptimized for new processors, which makes maintaining codebases difficult over time. Similar issues have long been addressed in the HPC community by libraries such as the Basic Linear Al-gebra Subroutines (BLAS) [2]. However, there is no analogous library for deep learning. Without such a library, researchers implementing deep learning work-loads on parallel processors must create and optimize their own implementations of the main computational kernels, and this work must be repeated as new parallel processors emerge. To address this problem, we have created a library similar in intent to BLAS, with optimized routines for deep learning workloads. Our imple-mentation contains routines for GPUs, and similarly to the BLAS library, could be implemented for other platforms. The library is easy to integrate into exist-ing frameworks, and provides optimized performance and memory usage. For example, integrating cuDNN into Caffe, a popular framework for convolutional networks, improves performance by 36 % on a standard model while also reduc-ing memory consumption.