This zen is running on 26 cores I think, so probably more like 10K to 20K
simulations per second.  Still, very impressive.

 

We might find that the stronger programs do fewer simulations per second,
since the recent growth in strength is mainly due to additional go
knowledge, which must make playouts slower.

 

Certainly Many Faces's playout per second on 19x19 is much lower today (at 2
dan, 2K PPS per core) than it was in 2008 (at 2 kyu, 6K PPS per core).

 

David

 

From: computer-go-boun...@dvandva.org
[mailto:computer-go-boun...@dvandva.org] On Behalf Of fastgo
Sent: Friday, June 03, 2011 5:58 AM
To: computer-go@dvandva.org
Subject: Re: [Computer-go] Zen19D is blitz 5d in KGS

 

It is really amazing that on 19x19 with less than 1M simulations, it can
reach 5D/6D level. Assuming 2K simulations/s, 15 seconds per move and 26
cores.

I cannot even imagine a MC program could reach that level on 9x9 with only
1M simulations two years back.

I would have to guess besides the smart logic in the playout side, there
must be smarter logic on the tree part.

If possible, it would be good to see some data from the strong bots how much
each part contributes in the strength by turning off the heavy logic on the
other part.

On Fri, Jun 3, 2011 at 1:10 AM, Aja <ajahu...@gmail.com> wrote:

Zen has some code to handle very simple semeais in the playouts. But I
am not satisfied with the code.
To solve the semeai issue fully, we will need a better algorithm to
share the tree information with the simulation.

 

 Such "some code to handle very simple semeais in the playouts" is still
very impressive. It's a pity that Yamato won't publish any paper about Zen's
algorithms. But anyway it's clear that we need something new and BIG to
solve semeai, rather than heavy knowledge implementation.

Aja 

_______________________________________________
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

 

_______________________________________________
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to