Author: Robert Hyatt
Date: 15:26:27 05/27/99
Go up one level in this thread
On May 27, 1999 at 08:52:03, blass uri wrote: > >On May 27, 1999 at 08:02:59, Robert Hyatt wrote: > >>On May 27, 1999 at 03:06:39, Jouni Uski wrote: >> >>>On May 27, 1999 at 03:02:37, Ed Schröder wrote: >>> >>>>On May 27, 1999 at 02:50:08, Marcus Kaestner wrote: >>>> >>>>> >>>>>>Who cares, it will be a great show anyway. >>>>>> >>>>>>Ed Schroder >>>>> >>>>>ed, you´re right. >>>> >>>>And I expect that some of so called big-names will have a very hard >>>>time as anything can happen in this super strong tournament. IMO there >>>>are 7-10 programs who can win this tournament. >>>> >>>>When I voted on CCC that was difficult. In the end I picked: Chess Tiger, >>>>Rebel (why not), Dark Thought, Conners and Zugzwang but I could have add >>>>many more. >>>> >>> >>>Never heard about Chessbase engines: Fritz,Hiarcs,Junior + multiple prosessors? >>> >>>Jouni >> >>It is not very 'clear' but the famous 'grapevine' would say these parallel >>engines are not very effective. And this can also be predicted because of the >>machines they are using. IE who would use a dual-cpu machine when quads are >>readily available... _unless_ you are doing something pretty ugly in the >>parallel search world: splitting only at the root. Just idle speculation since >>we won't ever hear of any 'good' algorithm of course. But I have heard that >>at least one of those is only getting 1.3X faster with 2 processors. > >How much faster does crafty get with 2 processors? > >Is there an overlapping when both processors analyze the same position in >different move order without knowing it? > >Do the 2 processors work all the time or maybe in part of the time >one processor must wait? > >Uri I haven't run such a test in a while, but a good guess is 1.7X. IE, as I have said before, I lose about 30% of each processor after the first, at least thru four which is all I have had time to test so far. If you split at the root, 1.3X sounds about right with 2 cpus. With 4 cpus, maybe 1.4X. With 4, I am hitting roughly 3.2-3.3 at present... There is some overlap.. which is where that 30% gets lost... and there is never any time where processors are 'waiting' in my code... the only lost time is doing duplicated work at times...
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.