Darren,

> Do you have more details

x86 64 bits, 32 GB ram, about 1500 playouts/s/core at the beginning
of a 19x19 game.

> effiency (actual playouts/second/core compared to pachi running
> on just one of the cores)

I have not measured it but it should be close to 100% with your definition
of efficiency since the tree is not locked. However I think that this
definition
is not the most useful one since it doesn't say anything about playing
strength.

Pasky and I are preparing a new paper which includes a section on
scalability
(both on a single machine and in distributed mode), which should answer
your questions. In short scalability on a single machine is very good and
scalability on multiple machines is hard but we are making progress there.
In the past I couldn't get performance improvements beyond 10 machines.
Now I can get improvements in playing strength up to 64 machines but 128
machines don't perform better than 64.

Jean-loup
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to