Re: Parallel processing

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

We need to find a tactical search algorithm that uses local
information.  I'm imagining some storage, a state machine,
and perhaps a few small adders at each point on the board to
detect local features, and evaluate moves for move ordering, plus
a central state machine to run the alpha-beta algorithm.  It can't
be too complicated since there are only a few thousand gates per
board point.  By doing move ordering operations in parallel on all points,
at 200 Mhz, there should be a huge speedup in tactics.  I think
one person could do it, if he understood hardware design.

David


At 12:39 PM 3/20/99, Darren Cook wrote:
>>I'd like to use the new 1 million gate Xilinx gate array to do some
>>really fast custom hardware for go.  If we put 4 or 8 of these on a PCI
>>card, we can get parallel processing for go in a very small box.
>
>I had a look at http://www.xilinx.com and it looks interesting, but it
>looks like a new set of jargon to learn.
>
>What is involved in moving tactical search C code into hardware? The
>tactical searches are the bottleneck in most programs, AFAIK. Is this the
>kind of project a single programmer can tackle, or is it more a job for a
>team of experts?
>
>Darren
>
>