Author: Robert Hyatt
Date: 20:07:10 11/09/98
Go up one level in this thread
On November 09, 1998 at 21:06:51, David Blackman wrote: >On November 09, 1998 at 20:01:50, jonathan Baxter wrote: > >>These famous "chess chips" of DB: did they do a full-blown alpha-beta search >>or did they just do static eval? Seems to me if they just do static eval >>then a plug-in card based on them for a PC is going to have a huge bottleneck >>just getting the positions into the chips. >> >>Jon > >They did 3 to 4 plies of alpha-beta (without null-move, and with a quiescence >search) and then a static eval at the leaves. The chips don't have a >transposition table but do have some sort of mechanism for detecting >repetitions. Eval weights can be uploaded to the chip from the host. It's not >clear to me if search parameters can also be uploaded. > >Given that each CPU of the SP is slower for this kind of thing than a modern PC, >and the SP ran 16 chess chips per CPU, i think a PC could run quite a few chess >chips flat out, if they were on a well built PCI card. Unless something changed that I don't know about, they did do a transposition table... IE it was originally a re-do of the Belle chess machine, which did a hardware transposition table. I don't remember the details of what goes on, ie perhaps the 8 processors on a single board have a shared transposition table. Will try to find out... Bob
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.