I have some experience with supercomputers… which I will just call the fastest computers at any given time. My experience is that software developers -algorithm developers, really - are generally far behind in exploiting the power that the hardware offers. The typical case that I used to hear was “you know that problem that used to take a day to solve? Well we can now solve it in a minute. Cool huh?”
Well actually I could have waited a day so … big deal. I want to solve NEW problems.
I’m well versed in numerical design so I understand that I need to parameterize a design in several dimensions and then search over that multidimensional design space for some “best” solution, testing with some merit function thousands of candidate designs. So faster is better but only if the damn algorithm developers would improve the multidimensional search algorithms to deal with millions of candidate designs smartly. But sadly, it will take years after the hardware is built to fully exploit the power of the new computers.
I’d be interested in counter examples or similar examples.