Re: computer-go: question -- Oracle of Life and Death FPGA

["Clay ChipSmith" <weiqi@xxxxxxxxxxxx>]

"computer-go" had a discussion some months ago about hardware accelerators.
Various aspects of the discussion took residence with me.  One, general
purpose hardware was expensive, giving acceleration only proportional to
hardware replication.  Two, special purpose hardware is also expensive, but
gave acceleration well beyond where general purpose hardware realized.
Particularly, an fpga may be arranged for specialized tasks.  The attached
comments give rise to wish to assign a task for such specialization, being
Oracle of Life and Death.

Now begs the question, is there hope this wish may be plausible?  Honestly,
I do not know in a certain way. Yet my sense tells me the realizable
parallelism in a single instance of board evalution may be sufficient to
make this wish plausible.  Move generation is a necessary part of the
acceleration process, and what little sense about move generation is
negative toward this wish, as the data volume of the patterns may be too
much to fit locally.  Within a short time, though, this pattern size problem
will be solved by Moore's Law.  So, here too there is hope this wish is
plausible.

Best Wishes, Clay ChipSmith ><>

-----Original Message-----
From: David Fotland <fotland@xxxxxxxxxxxxxxxxx>
To: computer-go@xxxxxxxxxxxxxxxxx
Date: Friday, December 10, 1999 9:08 PM
Subject: Re: computer-go: question


>A lot to tremendously, say 4.5 on your scale.  It would take a little
>bit of adjustment to the program to take advantage of this information
>and get that result.  Many Faces loses many games because it fails
>to live somewhere or lets some group die.  This oracle would free up
>my time to work more on territory evaluation.  I'd guess an immediate
>2 or 3 stone improvement, and another stone or two after a few months
>work, say to around igs 7k-9k*.  The fast oracle would let me do
>much more full board reading, since it would probably speed up the
>evaluation function by a factor of 5 to 10X
>
>Further improvement would need much better full board move generators,
>better evaluation of thickness, influence, etc.
>
>David
>
>At 05:46 PM 12/10/99 -0500, Tim Klinger wrote:
>>
>>I have a question for those of you with working go-playing programs.
>>
>>If you had an oracle that could tell you the life and death status of any
>>block on the board, taking essentially no time to do so, how much do you
>>think it would improve your program?  If you like, you can also assume
that
>>the oracle tells you how to go about making life for a block (if it's
>>possible).  You can even assume it tells you the *best* way to go about
it,
>>with respect to an evaluation function of your choosing.
>>
>>If it's easier, answer the question using one of the following ratings:
>>
>>(1) not at all, (2) a little, (3) a fair bit, (4) a lot, (5) tremendously
>>
>>and feel free to offer any qualifications.
>>
>>Thanks.  I'm looking forward to your responses.
>>
>>Tim
>>
>>Just in case there's any confusion about what I mean by a 'block', I mean
a
>>maximal set of adjacent stones of the same color (black or white).  I
don't
>>mean groups or chains or anything less solidly connected.