--- Tom McCabe <[EMAIL PROTECTED]> wrote:

> You cannot get large amounts of computing power simply
> by hooking up a hundred thousand PCs for problems that
> are not easily parallelized, because you very quickly
> run into bandwidth limitations even with gigabit
> Ethernet. Parts of the brain are constantly
> communicating with one another; I would be very
> surprised if you could split up the brain effectively
> enough to be able to both run one tiny piece on a PC
> and have the PCs communicate effectively in realtime.
> 
>  - Tom

It is not that hard, really.  Each of the 10^5 PCs simulates about 10 mm^3 of
brain tissue.  Axon diameter varies but is typically 1-2 microns.  This means
each bit of brain tissue has at most on the order of 10^7 inputs and outputs,
each carrying 10 bits per second of information, or 100 Mb/s.  This was barely
within Google's network capacity in 2000, and probably well within it now.
http://en.wikipedia.org/wiki/Google_platform

I think individuals and small groups trying to build AGI will have a hard time
competing with Google due to the cost of hardware.  It costs $2 million/month
just for electricity for their server farms.  Google is building a
supercomputer in Oregon that will have cooling towers 4 stories high.
http://en.wikipedia.org/wiki/Project_02

> 
> --- Matt Mahoney <[EMAIL PROTECTED]> wrote:
> 
> > --- Eugen Leitl <[EMAIL PROTECTED]> wrote:
> > > > Google already have enough computing problem to
> > do a crude simulation of a
> > [human brain]
> > > 
> > > Um, no. It takes 64 kNodes of Blue Gene/L to do
> > about 8 1/10th speed
> > > crudely-simulated
> > > mice, or about one realtime cartoon mouse
> > (assuming, the code would scale,
> > > which it
> > > wouldn't).
> > 
> > The Blue Gene/L simulation is at a lower level than
> > is needed to do useful AI.
> >  You don't need millisecond resolution.  In most
> > neural models, the important
> > signal is the rate of firing, not the individual
> > pulses.  I realize there are
> > exceptions, such as the transmission of phase
> > information for stereoscopic
> > sound perception up to 1500 Hz.  But for most
> > purposes, neurons have an
> > information rate of about 10 bits per second.  This
> > was measured in tactile
> > sensation in the finger.  (Sorry I don't have the
> > references).  In any case,
> > 0.1 seconds is about the smallest perceptable time
> > unit in humans.
> > 
> > The human brain has about 10^11 neurons with 10^4
> > synapses each.  Each synapse
> > represents about 1 bit of memory (Hopfield model). 
> > Therefore you need 10^15
> > bits of memory and 10^16 operations per second.
> > 
> > Google has about 10^5 PCs with 2-4 GB memory each,
> > connected by a high speed
> > Ethernet.  Therefore they have enough memory.  They
> > have a mix of different
> > processors, but a modern processor can execute 10^10
> > to 10^11 16-bit
> > multiply-add instructions per second using SIMD
> > (SSE2) instructions (more if
> > you add a GPU).  Therefore they have enough
> > computing power, or at least close
> > enough to do useful experiments.
> > 
> > 
> > 
> > -- Matt Mahoney, [EMAIL PROTECTED]


-- Matt Mahoney, [EMAIL PROTECTED]

-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=4007604&user_secret=8eb45b07

Reply via email to