I found the breakout sessions in the afternoon particularly fascinating for their paradigm-breaking look into the future of high performance computing. While FLOPS become free, huge scale and ultra-low-power devices (little voltage difference to identify bits) make error handling a topic of major concern, and the idea of “wasting” processors to perform redundant calculations to avoid errors becomes quite serious. Other ideas, like moving computation around the memory instead of the reverse, the rise of large data and questioning the role of GPU and other novel architecture was also an important theme. Breaking out of the cycle of simply scaling up the MPI paradigm will open new questions but require new hardware, new algorithms, and new kinds of thinking.