Paper on energy per instruction - using an appropriate numerical type

Several years ago I Read an excellent paper which discussed the knee jerk reaction of using double precision everywhere for scientific computation. The paper analysed the energy use per instruction.
As I remember the paper was written by someone high up in one of the big semiconductor companies. I could be wrong there.
I often want to cite this paper in discussions.
Can anyoen help please with a reference?

IO also may have asked this before on here. I have the memory of a goldfish.

1 Like

Are you maybe referring to John Gustafson and his work on unums, eg his book “The End of Error”?

No, not unums as I recall.