Author: Komputer Korner
Date: 01:39:03 12/18/98
Go up one level in this thread
On December 17, 1998 at 14:30:59, José de Jesús García Ruvalcaba wrote: >On December 17, 1998 at 04:08:31, Komputer Korner wrote: > >>On December 16, 1998 at 18:00:39, José de Jesús García Ruvalcaba wrote: >> >>>On December 16, 1998 at 16:42:57, Komputer Korner wrote: >>> >>>>On December 16, 1998 at 12:50:36, José de Jesús García Ruvalcaba wrote: >>>> >>>>>On December 15, 1998 at 02:08:19, Steven Juchnowski wrote: >>>>> >>>>>>Is there an explaination as to why the relative strength >>>>>>of a Chessmaster personality should vary under different time >>>>>>controls? >>>>>> >>>>>>Chessmaster 6555 appears to be stronger than Chessmaster 6000-4 >>>>>>under blitz conditions, why is this not necessarily true under longer >>>>>>time controls. After all both personaliies are using the same chess engine. >>>>>> >>>>>>Regards to all you Chessmaster reseachers. >>>>>>Steven >>>>> >>>>> That can happen with any two engines. If you slow down a current >>computer by a >>>>>factor of 1000 and play a match between a good current program on it against a >>>>>good old program in and old computer (from when the computers were 1000 times >>>>>slower) at standard time control; I am sure the old program will win, because >>it >>>>>was optimized to get the most of the available resources at that time (which >>>>>were scarce). >>>>> On the other hand, if somehow you speed up the old machine 1000 times >>and play >>>>>the match, then the new program in the new machine will win. Slowing down or >>>>>speeding up are equivalent to changing the time control, and now it is clear >>>>>that different engines are best at different time controls. >>>> >>>>At short time controls where hash tables are involved this is not true. Your >>>>time handicapping is thrown out of kilter by instant hash reads. The KK Kup is >>>>fair because even though it has time handicapping the hash tables get filled up >>>>quickly with the extremely long time controls and the hash reads are a very >>>>minor factor. >>>>-- >>>>Komputer Korner >>> >>> I did not mean time handicapping as is usually understood (i.e. giving >>>more time to one player). I meant hypothetical machines, one which is like a >>>modern PC but many times slower for one example, other one which is like an old >>>vacuum tubes computer but many times faster for the other. >>> >>> The difference is in the pondering time: giving more time for one player >>>gives also more time to the opponent to ponder. >>> >>> Also, I meant standard time controls, say 40 moves in two hours. Correct >>>me if I am wrong, I understand that you think that a modern program on a modern >>>machine at a very fast time control would play better than an old program on an >>>old machine at a slow time control. Unfortunately it is very difficult to check >>>this out: it requires an old computer with a chess program, and there is the >>>technical difficulty with the pondering time. >> >>The reason that the KK Kup is fair is that there is no pondering and the hash >>tables get filled up quickly. Different machine speeds are a tricky business in >>any other setup other than KK Kup rules. Of course if you switch the computers >>half way through the match then that is fair but that is impossible to do when >>measuring dedicated machines vs software. The modern programs will of course >>play better because of improved algorithms. If you simply handicap via >>difference in machine speed that will not be enough even if you disable >>pondering. >>-- >>Komputer Korner > > No doubt that the KK Kup is fair, but again I am meaning very different time >controls. Also, the algorithms have improved. I understand that you think that >even disabling pondering a current program+machine with a very reduced amount of >time would play better than an old program+machine. > Let me refer to the old times when it was impossible to make a full-width >alpha-beta 3-ply search in 3 minutes. The programs of those times were >invariably Shannon type B. On those very slow machines they were better than a >Shannon type A program, due to the scarcity of resources. > When faster machines arrived, type A programs began to dominate the scenario. > My point is that while the algorithms have improved in many ways, they have >adapted to faster hardware in a no less important way. That may be but any time you give different time controls and/or have different machine speeds you will run into problems of fairness and bias in testing matches. CSTAL is an example of your Shannon B type but so far has yet to make the big breakthrough that you seem to think that is possible. Deep Blue has shown that it takes both speed and knowledge to beat the best GMs. However the kind of speed needed is just not possible with micros at the present time. Your love of old time programs is touching but everything including knowledge algorithms improves over time. -- Komputer Korner
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.