Author: Robert Hyatt
Date: 18:05:40 07/20/00
Go up one level in this thread
On July 20, 2000 at 18:36:25, Chris Carson wrote: >My intent is not to mis-quote. So please do not >mis-quote me either. Thanks! > >I think Dr Hyatt has proposed two tracks of thinking >to support a claim that 97 DB could beat the leading >programs on the fastest micro's 90% of the time (including >DJ 6 on 8x-700 HW). > >1. DT scored 90% against micros from 1988 to 1995. DB is > at least 100 times faster than DT and todays fastest > micro's have not closed the gap (P6-200 was released > in Nov 1997). > > Chris Carson believes this is not a good representation of the > DT record. Almost all the games DT played against the micro's > were in reality on 386/486 HW during 1988 and 1994. The only > 1995 data that is present for DT is the 95 WCCC which DT lost > to Fritz 3 on a P-90 and drew against Wchess on a P-90. Thus > the 90% claim does not apply to P-90's. Also no record exists > that DT ever played against P-133's or P6-200 (pentium pro's). > If anyone can produce DT records showing a 90% score of DT > against top-5 commercial programs on P-90 HW or faster, I would > consider this a reason to change my view. As I said, you continue to miss the point. DT showed up in 1988. Each year it got better (new chips. More chips. New search extensions. New evaluation ideas. etc.) It continued to get better, and _nothing_ says it improved at a less rapid rate than the micro programs. And they started out light-years faster than the micros. And from 1988 until the last DB version in 1997, they _continued_ to widen the gap each year in terms of speed. And since they were clearly better from 1988 to 1995, and since they widened the speed gap yearly through 1997, I don't see how they were going to get 'caught'... > > Point two that Chris Carson makes is that the 8x-700 HW that > DJ6 ran on is 99.2 times faster than the P-90 (that DT had > problems with) and 450 to 900 times as fast as the 386/486 > HW that DT had the 90% score against. Thus in the speed gap > is a lot closer than the Dr. Hyatt picture paints. The NPS > gap may be higher, but that is not a valid comparison. THe gap was _huge_ in 1995. It is _still_ huge today. DB2 is as much faster than DJ on the 8-way box as DT was faster than the best micro in 1995... > >2. Dr. Hyatt quotes the 36-4 record that the DB team got during a > lab experiment that included a single DB chip running at 1/10 > speed. Dr. Hyatt says that shows repetability for the 90% > claim. > > Chris Carson says no it does not. The HW used was a P-90 as > reported by Murry Campbell in a later speech. Murry Campbell > was not sure of the s/s, perhaps Fritz3, CM2100, Crafty and > um-I don't remember. Chris Carson believes the DB team did > the experiment, but as Murry said it was not scientific, more > of a curiosity. Since we do not know what SW/Version/conditions > the 36-4 result originated from, the validity of the measure has > to be questioned. Also, the P-90 HW is 99.2 times slower than > the 8x-700, so Chris Carson does not think that a 36-4 result > could be repeated by 97 DB against DJ 6. > > Chris Carson's final point is that the 97 DB TPR = 2862 is > 160 points higher than DJ 6 TPR of 2702. Thus according > to the ELO rating system (www.fide.com) DB would have a > 22% win expectancy. I am not saying that DB should not > be expected to win, just that I do not believe DB would > win 9 out of 10 times against DJ 6. Dr Hyatt says that > these were human vs computer TPR's and not valid for > for comp vs comp. I agree to an extent, however I think > that 90% is a bold claim and not based on valid data. > >DB and the DB team did a great job. I do not understand why >some people want to give it more credit than it deserves. It >deserves a lot, 1st get GM norm (DT), 1st to get GM performance >(DT), 1st to beat strongest human in 6 game match. I do not >take anything away. :) > >Great yes, beat DJ 6 on 8x-700 HW 90% of the time. No in >my opinion. > >Best Regards, >Chris Carson
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.