Every optimization conversation eventually hits a wall that has nothing to do with the code. You cut allocations, you batch the I/O, you move the hot loop into a tighter language, and the machine still gets warm and the power meter still climbs. That floor is not an engineering failure. It is thermodynamics, and it was worked out about a century before anyone wrote a line of software.
Here is the part that is actually useful to know.
The Law That Sets The Floor
The second law of thermodynamics says the total entropy of a closed system never decreases. In practical terms, no process that does useful work converts its input energy entirely into that work. Some fraction always leaves as heat into the surroundings, and that fraction is not a defect you can design around.
This is why perpetual motion machines fail, why a car engine tops out around a third of the fuel's energy, and why a data center's cooling bill is a line item rather than a rounding error. One statement covers all three, because the law does not care what the machine is made of.
Entropy Is Counting, Not Chaos
Most of us learned entropy as "disorder," which is a bad definition that causes confusion later. Entropy is a count. For any macroscopic state you can measure, some number of microscopic arrangements would produce exactly that state. Entropy is the logarithm of that number, written S = k ln W.
Once you frame it as counting, the second law stops being mysterious. A system moves through the arrangements available to it, and the arrangements that are vastly more numerous are the ones you find it in. Heat spreads out because there are overwhelmingly more ways for energy to be distributed than concentrated. No tendency toward chaos is required, just arithmetic on very large numbers.
That framing is also why Shannon entropy in information theory carries the same name. Both quantities count how many configurations are consistent with what you know.
Where The Heat Actually Comes From
For a computer specifically, the interesting result is Landauer's principle: erasing one bit of information has a minimum energy cost of kT ln 2, released as heat. Overwriting a register throws away the knowledge of what used to be there, and that lost information gets paid for somewhere.
Real hardware sits orders of magnitude above the Landauer limit, so this is not what is making your laptop hot today. What it establishes is that the floor exists and is not zero. Computation is a physical process, and information is physical.
What This Means For Real Systems
The practical version is boring and useful. Efficiency work is not about beating the limit, it is about closing the gap to it. Every layer between your algorithm and the silicon adds losses, and nearly all the wins available live in that gap rather than in the physics.
It also reframes the energy conversation around large models. The question is never whether the heat can be eliminated, only how much useful work you extract before it leaves. That is a ratio, and ratios can be improved.
If you want the full picture, all four laws, energy and work, entropy, and the real world applications, the complete guide is here: https://www.learnhowtoscience.com/thermodynamics-explained/
The two pieces above have their own pages as well, the second law at https://www.learnhowtoscience.com/thermodynamics-explained/second-law.php and entropy at https://www.learnhowtoscience.com/thermodynamics-explained/entropy.php
Top comments (0)