Re: [gnugo-devel] Re: [computer-go] SlugGo v.s. Many Faces

[David G Doshay <ddoshay@xxxxxxx>]

On Aug 26, 2004, at 2:47 AM, wagjan@xxxxxxxxxxxxxxxxx wrote:

> May I ask the averages branching factor at the 1st ply ? If it is
> significantly less then 72, then do other processors nothing ?
We have not yet gathered all the statistics to know the average
branching factor, but the maximum we have tried is 8.

Because we know that a branching factor of 8 uses 10 cpus
(one for the master process and one for the pool manager)
we divide the cluster into smaller groups and run many
games simultaneously. It turns out that the pool manager
is not very busy at all, so I should get around to writing it as
a thread running on the master cpu, but right now it is OK to
stay with a consistent programming model.

We have also run simulations with smaller branching factors.
We have not completed the analysis to determine the
effectiveness of adding the last few "low value" (by GNU Go)
moves to the candidate list, but we know that we have selected
moves from all positions in the list. We just have not quantified
it or inspected all the game records to determine how important
we think those moves are.

But if your concern is unused processors, in some ways you
are correct. Our basic architecture is that of a pool of cpus
ready to be put to work. This implies that there will be idle
cpus at various times in the simulation so that they are ready
when needed. If this seems counterintuitive ... oh well.

Cheers,
David



_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/