One of the best chips is bad at maths
I will share with you a promising innovation in chips. A MIT project that was funded by DARPA resulted in a chip …
capable of processing frames almost 100 times faster than a conventional processor restricted to doing correct math—while using less than 2 percent as much power
The chip was actually doing imprecise summations but it was enough to perform very well on some hard tasks which do not required precise calculations. That is exactly what one does not expect from a chip’s functioning!
You can find more detail in this MIT review.