X-Message-Number: 13724
From: 
Date: Sun, 14 May 2000 12:49:59 EDT
Subject: Computing power for brain simulation, Re to T. Donaldson

T. Donaldson said in message # 13717:

> To Yvan Bozzonetti:
> 
> Basically you have a major underestimate of what would really be
> required to make a computer simulate even one neuron... >>
>

Thomas:

The common estimate for a neuron simulation is 10 000 floating point
operation per second ( one flops = a multiplication on 64 bits). This is
good for neural network studies, not I think, for a real brain simulation.

I have suggested to represent neural activity on a grid, this is far more
power hungry. For 10^10 neurons in a brain, the computing power soars
to 10^24 flops or 10^14 flops per neuron (100 000 billions flops). If you
think this is a gross underestimate, then please give your own value...

At this scale, the largest supercomputer of today could simulate
only one neuron with great pain...There, a full electrochemical cell
activity is simulated, not a simple basic electrical property. That 
technology is the one exploited in theoretical physics to simulate
quantum fields. See for example latice gauge theories on 
http://xxx.lanl.gov . I think I hover estimate the power requirred, most
chemical reactions in a neuron are liked to the cell life, not its activity
as a brain element.

Yvan Bozzonetti.

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=13724