# Modern Computer Efficiency vs Modern Nervous Systems

I am currently working on an evolutionary system and most of what I have heard is that a computer like mine at the moment would be able to simulate a bee sized brain (not taking into account the time needed to evolve it)

What I am wondering about is what is the storage capacity/computations per second of a normal human brain compared to a modern computer?

Cheers Martin

Brains aren't built for FLOPS. In fact, compared to computers, we suck at them. I can't do a floating point add of the decimal representations of 32-bit (single-precision) numbers in a second or less. I probably couldn't work with the binary representations any faster. So, in terms of FLOPS, my guess is that a bound based on adds would be somewhere around .001 to .1 FLOPS, and if we include multiplies, then maybe less than $10^{-3}$ FLOPS.

The fairest comparison is to measure the brain's computing power in instructions per second, or IPS, which would be comparable to integer operations for a CPU. Estimates for the computing capacity of a computer vary (see Ars Technica quoting an article in Science, insideHPC quoting a researcher, a UAlberta professor of psychology doing a back-of-the-envelope calculation, and an Oracle-sponsored project). The range seems to be $10^{11}$ to $10^{18}$ IPS, omitting prefactors from the scientific notation.

As for the storage capacity of the brain, those same articles quote a range of around $10^{8}$ to $10^{15}$ bits of storage, depending on estimation method. The range of estimates for both processing power and storage tend to try to account for the fact that the brain and computer hardware are designed for fundamentally different sets of tasks. Regardless, for the quoted processing power and storage numbers, the human brain is the more efficient of the two "computers" on a power basis, requiring roughly 2000ish calories per day of sustenance (plus the right micronutrients), whereas a computer typically requires a power supply in tens to hundreds of watts.

• @GeoffOxberry: How do you justify that instructions per second are comparable to integer operations, but not floating point operations? – Paul Mar 13 '12 at 3:57
• I think that the FLOP argument was that the CPU operates using a single bit on/off representation of electrical signals, while the brain does some sort of fuzzy logic using chemical signals. From that point of view CPUs and brain are not comparable. – fcruz Mar 13 '12 at 11:01
• @Paul: I believe the argument being made in the sources above is what fcruz has said; neurons or electrical signals seem to be treated as one "instruction" each. – Geoff Oxberry Mar 13 '12 at 13:01

This has been discussed ad nauseam...

A good survey of different estimations can be found in Ray Kurzweil's "Singularity in Near" p89 Overall, 10^16 - 10^18 flops or bits seems to be the ballpark figure

The various estimations are based on functional neuron behavior * number of neurons, modeling the pattern recognition of the retina and extrapolating for size of brain, etc...

For a different take on simulating the intelligence of a brain, you can try "Jeff Hawkins - On Intelligence"

• Thats what I have read... and am implementing... that book is like my holy grail... thanks for the answer man :) – marscom Mar 17 '12 at 17:25