Author: Robert Hyatt
Date: 10:09:54 10/08/02
Go up one level in this thread
On October 08, 2002 at 10:38:23, Uri Blass wrote: >On October 08, 2002 at 10:30:16, Robert Hyatt wrote: > >>On October 08, 2002 at 03:58:13, Tom Kerrigan wrote: >> >>>On October 07, 2002 at 21:35:23, Robert Hyatt wrote: >>> >>>>Never said there was _not_ one. I said "there was no reasonably efficient >>>>software implementation" that could be used in an engine." Big difference. >>>>You _can_ emulate _anything_ in software, by definition. But that doesn't >>>>mean you can use it to play games... >>> >>>Is that what you said? What was the date? I'd like to look that thread up. >>> >>>Also, it doesn't mean you can use it to play games? Then what software >>>simulation was Joel playing? >>> >> >>Deep Blue 2 on deep blue 1 hardware, for one thing. The software eval was >>around >>for a long time, but it was not used to play gaime so far as I was ever told. >>It was used >>purely for tuning the eval by playing over a bunch of games, and >>"least-square-ing" >>the total error. >> >>I wouldn't begin to guess when I wrote that. However, since I have had the deep >>thought >>software evaluation tuning code on my ftp site for years, it was a long time ago >>to be >>sure... >> >>However, I don't think anyone would attempt to lash that up to a search and play >>chess, >>for the reasons I gave you a year or two ago. The software emulation would be >>way too >>slow. > >Not everybody is interested in generating >a stronger program. > >I believe that there are people who are >interested in emulating deep blue in software >not for being number 1. > >Uri >> What would you get by doing so? In 1986 we had the Cray version of Cray Blitz that searched around 400K nodes per second. We had the pure FORTRAN version that implemented everything in Cray Blitz using arrays (vectors) and it ran (searched) about 10 nodes per second on a vax 11/785. Of what use is the latter when trying to work on the former? A one second search on the Cray would take 40,000 seconds on the vax. That is 11 hours. To see how it would do on a real search, which would probably require 6 minutes on the cray, would turn into about a month of computing on the vax. Not much is going to be learned by such an "emulation" IMHO. And in fact, mistakes could easily be made as we did that year, by tuning the slow version only to discover that the tuning for the fast version had been hurt by doing that. >> >> >> >> >>>-Tom
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.