Intel Corp (INTC)

[chipguy]

Interesting comparison of MPU energy cost today

64 bit multiply-add - 200 pJ
read 64 bits from cache - 800 pJ
move 64 bits across chip - 2000 pJ
execute an instruction - 7500 pJ
read 64 bits from DRAM - 12000 pJ

Notice it costs 15X more energy to go to memory than
read data from cache. The execution of a multiply-
add instruction burns 97% of the energy in overhead
and 3% in the arithmetic circuits.

Also notice it is more energy efficient to redundantly
perform 9 separate multiply-add operations in different
locations across a chip than do it once in one location
and broadcast the result across the chip! (presuming
the operands were already widely distributed).

from IDF paper TCIS001, slide 19