# Re: BLUE GENE and confusion about flops and tops and whistles

From: Dani Eder (danielravennest@yahoo.com)
Date: Sat Apr 01 2006 - 06:35:32 MST

> Excuse me, let me be more clear. I am not looking
> for a lesson in mathematics. I am in fact asking a
> question for SL4. I want to know why Russell Wallace
> said the CNS does 1-15 exaflops. I want to know his
> source or reasoning behind the claim. Compared to
> everything else I have read it seems very liberal.
>
My estimate:

10^11 neurons x 10^4 synapses/neuron x 100 Hz average
neuron firing frequency = 10^17 synapse firings/sec.

There is uncertainty as to how much data a synapse
firing would equate to in an electronic architecture.
If the brain is sloppy and redundant it could be
less than 1 bit. If the exact timing and synapse
strength matter, and a neuron does internal stuff
more complicated than (sum the inputs and fire if
a theshhold is exceeded), then it could be more than
1 bit by a fair margin.

So I place a factor of 30 above and below the synapse
rate to make it: 3x10^15 to 3x10^18 bits/sec.

A common value for a floating point operation is a
single-precision floating point multiply (32 bits),
so the flop equivalent is 10^14 to 10^17
(100 Teraflop to 100 Petaflops).

A value of 1-15x10^18 flops implies 320-5000 bits
per synapse firing, which seems a lot to me.

DRN

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT