[fxtentacle]

The HN submission was initially called "advice on using GPU for deep learning" so I was hoping for optimization tips, too.