4 ms·
"CPU cycles" is not a very good metric, in my view. However, the brain is indeed much more efficient when comparing energy consumption considering that it uses
by entity345 8y ago
"CPU cycles" is not a very good metric, in my view.
However, the brain is indeed much more efficient when comparing energy consumption considering that it uses only about 20W or power.
- trashtester 8y agoActually, training AZ (5000 TPUS for 8 hours) requires 1.4e10 joules, assuming that each TPU needs 100watts. For comparison, the human brain operating training 4 hours per day every day for 14 years also uses 1.4e10 joules. So the energy efficiency is quite similar.
- entity345 8y agoThat's comparing apples and oranges. The brain works and learns 24/7. It needs 20W to function.
- trashtester 8y agoSo how do you see that changing my point, that the brain seems to need about the same amount of energy as AZ does?
- entity345 8y agoThe point is that it obviously does not.
- trashtester 8y agoI'm not sure what your claim is. Have you switched to the view that AZ is more efficient (by a factor of 6, if you assume 24/7 consumption), or do you still think that the brain is more efficient at learning GO? Edit: In case this is the cause for your objection: In the calculation for the energy needed for the human brain to master go (where I ended up with 14GJ=1.4*10^6 J), I only counted the consumption during the average 4 hours per day that one can expect someone to train. If you count all 24 hours, the 14GJ budget would be spent in 2-3 years. (Obviously, nobody can master GO in 2-3 years, regardless of effort.)