Author: Uri Blass
Date: 09:36:10 10/18/03
Go up one level in this thread
On October 18, 2003 at 12:29:38, James T. Walker wrote: >On October 18, 2003 at 11:15:19, Uri Blass wrote: > >>On October 18, 2003 at 11:10:33, Drexel,Michael wrote: >> >>>On October 18, 2003 at 02:54:37, Thoralf Karlsson wrote: >>> >>>> 2 Junior 8.0 256MB Athlon 1200 MHz, 2784 >>>>DpFrit7 A1200 24-16 Fritz 8 A1200 47.5-26.5 Hiarcs9 A1200 17-11 >>>>Shred6P A1200 25.5-16.5 Junior7 A1200 23.5-16.5 Ruffian A1200 21.5-18.5 >>>>Shre532 A1200 9-5 Gandal5 A1200 25.5-14.5 Shred5 K6-450 36-10 >>>>Frit532 K6450 42.5-9.5 Nimz732 K6450 48.5-13.5 Hiar732 K6450 31-9 >>>>Goliath K6450 8-1 Craf17.07 K62 13.5-4.5 >>>> >>>> 6 Deep Junior 8.0 256MB Athlon 1200 MHz, 2748 >>>>Shr7.04 A1200 22-28 Fritz 7 A1200 9-11 CT 15.0 A1200 23.5-16.5 >>>>Hiarcs8 A1200 23.5-20.5 CT15.0 K6450 26.5-15.5 DpFritz7 K6-2 26-16 >>>>Fritz6 K6-450 33-11 Junior6 K6450 33.5-14.5 Nimzo 8 K6450 8-5 >>>>SOS K6-2 450 39-6 Goliath K6450 33.5-10.5 >>>> >>>> 7 Hiarcs 9.0 256MB Athlon 1200 MHz, 2746 >>>>Shr7.04 A1200 7-8 Junior8 A1200 11-17 Fritz 8 A1200 5.5-8.5 >>>>Shredd7 K6450 9-5 Junior7 K6450 13-6 Fritz6 K6-450 8.5-4.5 >>>>Junior6 K6450 31.5-14.5 Shred5 K6-450 34-12 Hiar732 K6450 12-2 >>>> >>> >>>I do not see the point of testing a Deep version and a single version of a >>>program on a single processor. >>>IMO this is just a waste of time. >>>You could have tested a strong amateur engine instead. >>>Aristarch,List,Little Goliath or CLC for example. >> >>I agree that there is no reason to test both Deep version and single version >>when they are the same engine. >> >> >>> >>>>There are three new entrants in this edition of >>>>the SSDF rating list. Fortunately enough all three >>>>have received clearly higher ratings than their >>>>predecessors! >>>> >>>>The first program which we have tested is Amir Bans >>>>and Shay Bushinskys Deep Junior 8.0 on a single >>>>Athlon 1200 MHz. After 432 tournament games >>>>against other programs, it has now a SSDF-rating of >>>>2748. That is 49 points above it's predecessor >>>>Junior 7.0, which appeared somewhat more than two years >>>>ago. >>>> >>>>As perhaps can be understood from it's name, Deep Junior 8.0 >>>>is primarily intended for PCs with several processors. When >>>>used on a single processor, the search is said to be slowed >>>>down about 5% compared to Junior 8.0, which we also have >>>>tested. After 545 games Junior 8.0 Athlon 1200 MHz now >>>>has a rating of 2784, 85 points more than Junior 7.0! >>>> >>>>The present rating difference between the single and >>>>multiprocessor versions of Junior 8.0 is 36 points. The >>>>slight speed difference between them should give only >>>>3-4 rating points. If the programs otherwise play identically, >>>>and that is what I have understood, the rest must be >>>>attributed to statistical factors. >>> >>>No wonder >>> >>>Junior 8 did not play against Shredder 7.04 nor Shredder 7. >>>It did not play against a Chesstiger version too. >>> >>>Instead it played 60 games against Shredder 5/5.32, 52 games agsinst the old >>>Fritz 5.32 on inferior hardware and 62 games against the old Nimzo 7.32 >>>on inferior hardware. >>> >>>Michael >> >>Why do you think that playing against weaker opponents mean bigger rating? >> >>It is often not the case and I remember that Junior5 lost rating from playing >>against old Comet on inferior hardware. >> >>Uri > >Uri, you conitnue to bring up this single instance over and over. It may be >just an aberation and not normal. My experience tells me that computers do >exagerate their rating differences and playing weak opponents will on average >give a higher rating. More games vs top opponents will give a better indication >of Junior 8 rating. My blitz testing of Junior 8 shows it has only a 25 Elo >increase over Junior 7. The SSDF list shows twice that much improvement. Maybe the difference at long time control is bigger than the difference in blitz. It is possible to check all matches between program when the difference is at least 100 elo to see if the stronger software+hardware scored more than expected. I do not know of a proof of that claim based on investigation of the results. Uri
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.