Author: Vincent Diepeveen
Date: 16:16:44 09/20/00
Go up one level in this thread
On September 20, 2000 at 18:06:59, Roger D Davis wrote: >Hello. Lots of folks are now working on Peer-to-Peer networks to exploit the >trillions of unused CPU cycles going to waste on desktops everywhere. Intel is >using such a network to assist in CPU design. SETI@home is using something like >this to assist in its analyses. What is the speed of communication at those networks, so the delay the first few bytes get? >What about a Peer-to-Peer chess program? All that woudl be needed would be a >means of distributing the moves to the various CPUs on the network. Won't it >soon be possible to use this technique to search deeper than Deep Blue, maybe >even without null move? Remember, it could be possible to involve THOUSANDS and >THOUSANDS of desktop machines. Nullmove has pretty proven itself. And note that deep blue ran without it, so it searched way less deeply as nowadays nullmove programs. To get insight in how deep thought evaluation looked like get its source at tim mann's homepage! It's not exactly heavenly well. Especially its pawn terms pathetic, just like deep blue handled pawn structure pathetic, only a4 move against kasparov in game 6 was a big surprise for kasparov i bet. >Such a program could run full time on ICC, with machines joining and unjoining a Just starting such a program on icc would already forfeit it before it's booted. Apart from that there are other problems. Like diep's parallellism, which is shared memory, already suffers heavily if other processes are running on one of the cpu's. >dynamic network that would increase and decrease in strength throughout the day, >but probably always be stronger than than all but the best. The person who wrote >such a program might even challenge Kasparov to a match, and you can imagine >that maybe a hundred thousand people would volunteer their computers to >participate in such a match. but kasparov would win on time before even a 1 ply search is carried out! >This is just speculation, but what is really possible given Peer-to-Peer? >Roger Are you talking about tcp/ip over the internet, so communication speeds of say at minimum 200 milliseconds a message? Imagine you run on 10000 processors and the command is to stop the search, how long till all processors know the search is stopped? Exactly!
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.