IBM
<p>Aug 02, 2012 10:34 AM
 
Excellent paper.
 
they describe a Neural Tissue Simulator  implemented on a machine like Blue Gene/P
 
TODAY max size of Blue Gene is only abt 1/10 smaller than what is requrired to sim a human brain.
 
Furthermore, to achieve a FULL DAY of of sim. time  (they like one full day...
24 hours includes all the diurnal rhythms)  and to run it in 1 day to 1 week
would take only a 104 speed up  (ie an exaflop)  expected within one decade!!
 
CAVEAT: sims that explicitly model <b>molecular diffusion</b> or  <b>extracellular field potentials</b>
may have signif higher comp costs... not included in this estimate!!!
 
BUT (on the bright side):  a LESS costly modeling approach to explain global brain function
(ie more abstracted and less detailed sim)  may provide a theor. basis
for predictive whole brian sim  ... that would LOWER the computational requirements.</p>