Agreed for a supercomputer design it would be awful, but for a stereotypical 99.9999% CPU idle applications, the big win would be total cost of ownership / cape

Agreed for a supercomputer design it would be awful, but for a stereotypical 99.9999% CPU idle applications, the big win would be total cost of ownership / capex. On a long term view from many decades of observation, the digital world is bifurcating into one world thats 100% utilized all the time in data centers or real time control or video gaming, and another world that's 99.999% idle but when its active the users want it infinitely fast. And the two worlds don't talk or cross pollinate and are gradually using separate hardware. Having spent some time in electronics labs, given experimental results with smoke emitting diodes, and given the GPIO ports, there is minimization of the total cost of hardware experiments. If you assume 5% of boards will be destroyed per PID/motion control lab, then minimization of board cost is key to minimization of cost of running the lab. Implementing a PC that costs 20 times as much but provides 100 times the performance would only result in increasing the cost of the lab by 20 times. There is also a total wattage ecological argument. I have a pi running a 3-d printer and it works great and only draws a couple watts, but the web application is extremely fat and pointlessly featureful "because the cycles are there so may as well use them" and its not like installing a 100 watt PC would make better prints, it would just make an even more elaborate UI to use all available CPU cycles thus burn more coal. A 300 watt tor node doesn't really do anything a 3 watt tor node does, other than burn 100 times the coal.