Oh, I really like his analysis on "Parametric Reach" for the various architectures:
> If you want to commit to a single architecture, it’s important to know which one gives you the most headroom to move up. I created a fictious “times better” score by comparing the the part tested with the best part available in the same ecosystem — this usually means fairly comparable peripheral programming, along with identical development tools. I multiplied the core speed, package size, flash, and RAM capacities together, ratioed the two parts, and then took the quartic root. Essentially, if every parameter is double, it is considered “2.0 x” as powerful.
> If you want to commit to a single architecture, it’s important to know which one gives you the most headroom to move up. I created a fictious “times better” score by comparing the the part tested with the best part available in the same ecosystem — this usually means fairly comparable peripheral programming, along with identical development tools. I multiplied the core speed, package size, flash, and RAM capacities together, ratioed the two parts, and then took the quartic root. Essentially, if every parameter is double, it is considered “2.0 x” as powerful.