The problem is of course that very few pieces of software beyond the benchmarks actually can scale to 1.6 million cores, so I wonder what the actual functional

The problem is of course that very few pieces of software beyond the benchmarks actually can scale to 1.6 million cores, so I wonder what the actual functional speed-up is for most research codes moving from the past generation to this generation of hardware, and what is the general lag time to bring a code online on a machine like Sequoia that can take full advantage of the tech.