Not quite true. Bennett showed that the heat dissipation was associated with *forgetting*, not *computation* per se. If your computation is *reversible*, then it can be done with essentially zero dissipation. An interesting question is how Bennett's theorem applies to crypto. Crypto is mathematically reversible (at least deterministic crypto is) -- else you could never recover the plaintext, but usual implementations are not. So any proof of crypto hardness is going to have to deal with this issue; otherwise, there are going to be lingering possibilities of efficient breakage of the crypto. (Notice that Shor's quantum algorithm doesn't violate this principle -- it allows the computation to evolve unitarily, i.e., reversibly.) https://en.wikipedia.org/wiki/Reversible_computing At 07:36 PM 5/18/2016, Keith F. Lynch wrote:
The amount of computation you can do per unit energy is proportional to the absolute temperature of the computer.