Still, even number of playouts is not the end-all I believe. I have the
distinct impression that eight cores running for one second plays
considerably worse than one core running for six seconds, even though the
number of playouts is in the same ball-park. I haven't had the time to do an
extensive test on that yet but I'm convinced that the picture is more
complicated than just looking at total computing power.
Sure.
In 19x19 however, the results are suprisingly good (even without shared
memory) from the point of view of the equivalence between N-cores x 1s and
1-core x Ns.
But the main issue is more the fact that
whenever
"a parallel version wins 95% against the sequential counterpart",
as well as
"the version with N seconds wins 95% against the 1 sec. counterpart",
a human player equivalent to the strong version (i.e. 50% winning rate)
has a winning rate far from 95% against a human player equivalent to the
weak version.
Whatever maybe the time setting of e.g. MoGo, or whatever may be the
parallelization, MoGo plays the same
form of Go; i.e. games in which each player tries
to kill some groups which are obviously alive. The weakness remains,
but this particular form of Go is played much better by the parallel
version: it wins almost always against the other version. But against
humans, the difference is much smaller: the human will exploit the
weaknesses.
It's a bit tricky to explain :-)
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
