Training Benchmark for Deep neural networks

I read the TBD arXiv paper but I haven’t yet had time to write a summary.