Author: Robert Hyatt
Date: 10:15:13 05/02/01
Go up one level in this thread
On May 02, 2001 at 12:21:42, Uri Blass wrote: >On May 02, 2001 at 11:43:10, Robert Hyatt wrote: > >>On May 02, 2001 at 02:32:27, Chessfun wrote: >> >>> >>> >>>>A SMP version of Tiger exists. Actually existed last year. >>>>A friend of mine went on a long hollyday last year, and instead of leaving his >>>>dual system unused at home, he lent it to me. >>>>I developped a SMP version on top of Chess Tiger 12.2. >>>>But as I already told several times SMP was not a priority for me. So once I had >>>>SMP working, I disabled the code in Tiger's source code. I did not erase it of >>>>course, I just deactivated the relevant parts of code. >>>>Since then, Tiger has evolved on many points, including a different internal >>>>move coding and a different hash table structure. >>>>So now I cannot just reactivate the SMP code. It will not work directly. I need >>>>to reactivate it and adapt it to what has changed in the latest Tiger engines. >>> >>>>It's not a lot of work. The principle of the SMP algorithm does not have to be >>>>rewritten, it's just a matter of adapting the code. >>> >>>>I just asked for a little delay in order to do that and to double check that >>>>everything was working. >>> >>>>You can have your doubts about this SMP version, but after all if I send a SMP >>>>Tiger that crashes in Cadaques, I will be the one to look stupid, not the >>>>organizers of the Kramnik match. >> >> >> >>I'm willing to let a lot of hype slip by, but not _everything_. It seems you >>are saying you developed an SMP search over a weekend. I have too much >>experience with parallel search to believe that. I don't think I could even >>steal my SMP code and move it into something like gnuchess in a single weekend, >>and get it working reliably. >> >>Believe me, this is _not_ a weekend task. It is not a month task. It is really >>not a year task. >> >>I would not ever believe there is an SMP Rebel until there is a windows version >>of Rebel. DOS is _not_ capable of running on an SMP platform itself, much less >>managing threads in any way... I have a _really_ hard time believing that a >>robust SMP algorithm is a week-end project, unless I am a far worse programmer >>than I believe. > >It is also possible that there is another way to do programs parallel that other >did not think about and I do not know what is the speed improvement that >Christophe got from the parallel program. > >I understand that patzer used a parallel version of their program and it was not >hard for the programmer to do it parallel. >The speed improvement from 2 processors in that case was 1.2 > >The interesting question is what is the speed improvement that christophe got >from the parallel tiger12.2 > >Uri 1.2 I don't call a parallel search. that might mean 1.3x or 1.4x with 4 cpus? it would be better to spend the time cleaning up and optimizing the code, as the speedup would probably be more...
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.