Author: Keith Evans
Date: 16:08:05 02/28/02
Go up one level in this thread
On February 28, 2002 at 12:57:19, Robert Hyatt wrote: >On February 28, 2002 at 05:23:47, Gian-Carlo Pascutto wrote: > >>On February 27, 2002 at 19:59:35, Peter McKenzie wrote: >> >>>Anyone know anything about Brutus which is playing in Paderborn? I know it is >>>Chrilly Doningers new program, but not much else. Heard a rumour it has special >>>purpose hardware, can anyone confirm this? >> >>From Vincent Diepeveen (some stuff inbetween deleted): >> >>Diepeveen(FM) kibitzes: brutus is fpga chip >>Diepeveen(FM) kibitzes: in software 50-70k nps >>Diepeveen(FM) kibitzes: and then each node they get from fpga is a 2-3 ply >>Diepeveen(FM) kibitzes: with nullmove of course >>Diepeveen(FM) kibitzes: no hashtables on chip >>Diepeveen(FM) kibitzes: very little eval on chip because each logics circuit >>costs money >>Diepeveen(FM) kibitzes: donning searching 14 ply now >>Diepeveen(FM) kibitzes: but 14 thin plies >>Diepeveen(FM) kibitzes: it is a world achivement >>Diepeveen(FM) kibitzes: he made the entire chip himself >>Diepeveen(FM) kibitzes: which is an incredible achievement > >That last I don't understand. Ken Thompson did Belle by himself. Hsu put >Belle on a chip and improved it, by himself. Doing it a _third_ time is an >"incredible achievement"??? > >:) > > Well it at least shows that FPGAs have made significant progress since your 1997 statement: "The next level up is to put "belle on a chip" like Hsu did in the first DT chips, but that's a horse of a different color and is likely beyond what you want to attempt using FPGA technology." I've been waiting for somebody to do something like this in a Virtex class part - maybe even get some form of sponsorship from Xilinx. It will be interesting to hear about the details. I have protyped two ASICs running at speed in Virtex parts, so I don't see why you couldn't duplicate Hsu's chip. And a lot of my buddies at networking companies are making production units with 1000+ pin plus Virtex parts which have very fast memory interfaces - so hash tables would not be out of the question. 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.