DNNMark: A Deep Neural Network Benchmark Suite for GPUs
DNNMark is a highly configurable, extensible, and flexible benchmark suite consisting of a rich set of Deep Neural Network primitives as benchmarks. It not only provides individual benchmarks of DNN layers such as convolution, pooling, LRN, and etc., but also allow users to build their own preferred complicated models with combinations of different layers for GPU-specific benchmarking. All the DNN primitives are implemented using cuDNN and cuBlas library and tested on Nvidia K40 with CUDA 8.0.
DNNMark is available HERE.