Asklemmy

There is no real comparison with present computers like this. Humans are massively parallel in operations. Present CPU architectures need the speed because they handle a tiny amount of data per cycle.

The primary bottleneck is the L2 to L1 cache bus width. This is one of the main bottlenecks that makes the CPU perform poorly with large language models.

If you remove such bottlenecks and let everything happen at once and start using analogue computational elements for neural networks, the needs shift entirely. The speed requirement is specific to the technology/efficiency/scale. The human brain is orders of magnitude more energy efficient than silicon. It has its issues, but it has too many advantages for sustainability long term. It is the ultimate self replicating recyclable machine.

In a properly designed vivarium, a system could only require cyclical sunlight for powering a self sustained computational unit on geological time scales, assuming all respective lifeforms were fully understood scientifically.