Author: Vincent Diepeveen
Date: 09:31:03 05/26/05
Go up one level in this thread
On May 26, 2005 at 12:17:11, Uri Blass wrote: >On May 26, 2005 at 12:02:37, Vincent Diepeveen wrote: > >>On May 26, 2005 at 10:40:04, Darrel Briley wrote: >> >>>There's another article on Chessbase site today concerning the upcoming match. >>> >>>http://www.chessbase.com/newsdetail.asp?newsid=2412 >>> >>>It's interesting to note that despite the upcoming Adams/Leko match beginning >>>June 2nd (a rapid match), the match with Hydra is getting all the press. Some >>>highlights from the article: >>> >>>...processing power equivalent to more than 200 standard PCs >> >>32 pc processors >> >>Please prove me that such a simple program like hydra cannot search with >>millions of nodes a second single cpu in software. Even crafty can get what is >>it 2+ million nps at an opteron or so? > >How do you know if hydra is simple or not simple? It is not easy to make a simple software program in hardware. Nor is it easy to tune all that. But yes it has a very limited set of knowledge, Chrilly never made a secret out of that. In fact in the games it plays you can easily deduce this if you study them. > >If you want to compare between programs I think that it may be more interesting >if you compare Crafty with other free source code programs like >Fruit-Toga or Glaurung. Fruit nor Crafty are playing at a 32 processor machine against Adams, so definitely Hydra is more interesting to speak about and whether Adams will be motivated to play well. I know how hard it is for persons with your personality to be able to realize with common sense the imporance of a game you practice at home against Fruit versus the importance of a match Adams versus Hydra, but take it from me that Hydra versus Adams for mankind minus 1 is more interesting to speak about. Comparing fruit/crafty with hydra is an insult for Hydra nevertheless. It's far better and especially more agressive tuned that hydra. Vincent >Uri > >> >>Additional it's without hashtables last X plies and very poor move ordering >>techniques in hardware (it does have killer moves as opposed to deep blue, so >>it's *far more* efficient in hardware than DB was at its days). >> >>>...64-way cluster computer >> >>32 processors used though. I'm sure it would be more effective when run in >>software. However would you have written this posting then? >> >>>...16 nodes of four computers, with each node boasting 32GB of memory >> >>Not used last 6 plies or so in search. >> >>>...Each computer has an Intel Xeon 3.06 Ghz. >> >>Irrelevant, search is in hardware. >> >>>...It has never been beaten by a human. >> >>Oh there is *many* programs that can say that with respect to official matches. >> >>>And some of the expamples provided to demonstrate Hydra's power: >> >>>-1 second to analyse 200 million chess moves and chose the best one. This >> >>Please note the node count of hydra is just guessed. There is not hardware >>beancounter. So he's just doing software nodes times some optimistic number of >>nodes a search which is guessed. And no doubt a factor 3 to 4 too much. >> >>>includes projecting the game 18-40 moves head (6 more than Deep Blue) >> >>Deep blue searched 10 ply search depth in crucial positions and up to 12 ply >>deep in other positions. Hydra will get double that depth. So that's 10 moves >>look ahead. >> >>>-1 millisecond to calculate all possible angles and determine whether Luis >>>Garcia’s shot was a goal in the Champions League semi-final >>>-1 second to match a finger print to any person within the UK >> >>No way, in any arabic state they're not clever enough to develop such techniques >>themselves :) >> >>>-Less than 1.5 minutes to match a finger print to any person in the world >>>-Approximately five minutes to calculate every prime number between 1 and 1x1051 >>>(a sexdicillion) >> >>Hello, i can calculate prime numbers up to 10 million digits at my pc nearly, >>though not within 5 minutes. >> >>A prime number of 1051 digits is really peanuts to calculate within seconds. >> >>I'll have to refer to some other software mine there. >> >>Vincent
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.