Strangely, Matrix Multiplications on GPUs Run Faster When Given "Predictable" Data! [short]

Great minds discuss flops per watt.

Thonk From First Principles