Author: Keith Evans
Date: 10:58:32 04/03/02
Go up one level in this thread
On April 02, 2002 at 23:41:05, Robert Hyatt wrote: > >This depends on what is implemented. IE I suggested including a FPGA >MakeMove() and UnmakeMove() function to take care of the hash updates. I think that there's some fundamental confusion here. There are two things that we're talking about. One is Slate's potential project and the suggestions that you have made, and the other is what Marc Boule has actually done. Slate was hoping that Boule's work would be sufficient, and I was basically on your side arguing about the limitations of that approach. And by the way Marc himself totally agrees with you. > >But in the case of "chiptest" it implemented a full alpha/beta tree search, >which means that if this became an FPGA process, you could give the FPGA a >position to search and a depth, and sit and wait for milliseconds while it >does the search. That would result in minimal PCI traffic... and let the >hardware run at full FPGA speed, which might be well beyond 30M nodes per >second easily... ChipTest the chip didn't implement alpha/beta search or evaluation. That was done externally. If you reread what I wrote I am referring to the approach where just the move generator is put on the PCI bus, and why you won't get much (if any) acceleration by doing that. I understand the basic API for a Deep Blue style chess chip and why PCI bandwidth would not be an issue for that. Regards, Keith
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.