Author: Omid David Tabibi
Date: 15:36:29 09/09/04
Go up one level in this thread
On September 09, 2004 at 18:03:50, Matthew Hull wrote: >On September 09, 2004 at 17:50:06, Omid David Tabibi wrote: > >>On September 09, 2004 at 17:37:04, Matthew Hull wrote: >> >>>On September 09, 2004 at 17:25:54, Omid David Tabibi wrote: >>> >>>>On September 09, 2004 at 17:17:49, Matthew Hull wrote: >>>> >>>>>On September 09, 2004 at 17:11:35, Omid David Tabibi wrote: >>>>> >>>>>>On September 09, 2004 at 15:20:49, Matthew Hull wrote: >>>>>> >>>>>>>On September 09, 2004 at 15:13:25, Omid David Tabibi wrote: >>>>>>> >>>>>>>>On September 08, 2004 at 20:52:31, Dan Honeycutt wrote: >>>>>>>> >>>>>>>>>On September 08, 2004 at 14:43:43, Omid David Tabibi wrote: >>>>>>>>> >>>>>>>>>> >>>>>>>>>>The average level of discussion in WCCC is several hundred Elo above that of >>>>>>>>>>CCC. >>>>>>>>>> >>>>>>>>> >>>>>>>>>That's logical since everyone there is an experienced chess programmer whereas >>>>>>>>>CCC caters to all levels. However, are there topics that you would be willing >>>>>>>>>to discuss at WCCC that you would not be willing to discuss at CCC? >>>>>>>> >>>>>>>>Most of the discussions here are relevant to under 2300 rated programs, and >>>>>>>>rarely any ideas of further use are discussed here. In WCCC when you play versus >>>>>>>>another engine, you always engage in some form of discusson with the programmer >>>>>>>>sitting on the other side of the table, or when you are in a restaurant you also >>>>>>>>speak about computer-chess issues with other programmers, etc. Of course most >>>>>>>>programmers don't discuss the details of their "weapon of mass destruction", but >>>>>>>>the discussion level is much higher, and many subjects that are brought up there >>>>>>>>never appear on CCC. >>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>>>I don't understand what your question has to do with the post you replied to, >>>>>>but here are the answers anyway: >>>>> >>>>> >>>>> >>>>>I was just trying to elevate the level of discussion on CCC. >>>>> >>>>>:) >>>>> >>>>> >>>>> >>>>>> >>>>>>>So, to how many processors have you scaled Falcon? >>>>>> >>>>>>Currently 4 processors is the maximum. >>>>>> >>>>>>>Will it run on 8+ NUMA machine? >>>>>> >>>>>>No, I don't have acces to such a machine. >>>>>> >>>>>>>Is it portable (like GNU or Crafty) or is it limited to Windows? >>>>>> >>>>>>Yes. Most of my parallel tests are on Linux. >>>>>> >>>>>>Any more questions? :) >>>>> >>>>> >>>>> >>>>>Can you post speedup data? We know what Bob has reported. >>>>> >>>>>Also, is speedup the same for SMP and NUMA? >>>> >>>>I told you, I don't have access to NUMA machine. >>> >>> >>>Can you post speedup numbers? Perhaps some logs? >>> >>>What strategy do you use to split the search? Were there any papers that were >>>of help to you in the development process? >> >>Is it an interrogation of some kind? :) > > >You asked earlier if their were any more questions. :) > >Not many amateurs (or even some pros) are attempting parrallel search. Also, >not everyone uses the same strategy, for example: threads (Crafty) versus >processes (Sjeng, DIEP). Falcon uses multiprocessing. > >I think a lot of people are interested in parrallel efficiency. Some have >posted their data on this. Are you still working on this aspect? Yes. > >Do you use the ICC compiler for Linux or some other? On Linux I use GCC. > >Do you feel that you have some methods that are of your own discovery, and so >perhaps not used in other programs? We all suspect there are some secrets in >commercial programs. However, I wonder if their real "secret" is that they are >more harmoniously tuned in the evaluation. It seems that this is a major >benefit that takes a great deal of effort to achieve.
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.