J Storrs Hall, PhD. writes:
As long as the trumpets are blaring, Beyond AI is coming out this month,
with
the coolest cover I've seen on any non-fiction book (he says modestly):
http://www.amazon.com/Beyond-AI-Creating-Conscience-Machine/dp/1591025117
Cool! I just pre-ordered my copy!
Look at Brook (http://graphics.stanford.edu/projects/brookgpu/) ... and
GPGPU in general (http://www.gpgpu.org/cgi-bin/blosxom.cgi).
I'm also just beginning my experimentation with modern hardware, and just
got
a new machine with two Nvidia 8800GTX boards. That G80 architecture is
moving
explicitly to a GPGPU architecture (by which I mean it doesn't have separate
vertex and pixel processors, just 128 general-purpose processors per card.
They
have some pretty decent programming tools for it (called "CUDA").
If you want to use the built-in SIMD instructions in the X8x architecture,
there are versions of BLAS that support them: both AMD and Intel have
native versions for download
If you are working in a somewhat low-level language and don't mind a little
bit
of effort, you can embed the assembly directly to use the scalar functions.
To get my feet wet with this, I just wrote a mandelbrot set exploration
program
that does this and it's amazing how far things have come recently. The CPU
on
my new machine is an intel quad core at 2.7 ghz. With each one executing a
4-wide
simd instruction (single precision), that adds up to 43 gflops peak, which
isn't anywhere
near the peak of the graphics cards but isn't too shabby.
As I just start to work on some AGI-type stuff myself, one of my "premises"
is
that it pays to think about models that lend themselves to efficient
implementation
on available hardware, in direct opposition to YKY's recent post on that
subject.
-----
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=231415&user_secret=fabd7936
