Author: Hristo
Date: 01:01:25 04/12/02
Go up one level in this thread
On April 12, 2002 at 03:24:59, Slater Wold wrote: >Just came back from a 2 week vacation, and boy, it was nice. Also saw Marc >Boulé found his way here; Welcome! > >Now, a DB rant: > >There are at LEAST 100 reasons I can think of why Kramnick would make such a >ridiculous statement. Here are a few: > >1.) to boost interest in the game > >2.) to boost his own "man vs machine" repitore. example: "Kaspy couldn't defend >man against the machine, I CAN!" > >3.) refer to #2 and repeat, "..I can beat Fritz 7, and thus beat DB, and invite >you to pay me $1,000,000 to beat your program! (And a small settlement if I >lose).." > >Fritz 7 is a good program, without a doubt, one of the best for PCs. HOWEVER, >get real. 4M "wiser" nps isn't CLOSE to 200M "wise" nps. No matter how you >look at it. > >Crafty 17.07 is a good starting point. This is an older program, with a more >basic search/eval, as compared to Fritz 7. Crafty 17.07 (K6-2 450Mhz) now holds >a 2490 ELO on the SSDF list, while Fritz 7 (Athlon 1200mhz) has a 2748. Pretty >big difference. > >Lets compare numbers: > >First we need to see what Crafty 17.07 would be rated running on the same >computer as Fritz 7. We'll take the "generally" accepted rule that doubling Mhz >is equal to 50 Elo. (Not exactly, but close.) > How close before your whole theory goes to hell? ;-) Cheers, hristo >Therefore we can estimate that Crafty 17.07 on an AMD Athlon 1200 would be rated >(closely) about 2550 Elo. 198 Elo lower than Fritz 7 on the same HW. > >Crafty 17.07 benchmark on an AMD 1200 mhz: > >EPD Kit revision date: 1996.04.21 >unable to open book file [./book.bin]. >book is disabled >unable to open book file [./books.bin]. > >Crafty v17.7 > >White(1): bench >Running benchmark. . . >...... >Total nodes: 81501850 >Raw nodes per second: 696597 >Total elapsed time: 117 >SMP time-to-ply measurement: 5.470085 >White(1): > >Roughly 700k nps = 2550 Elo > >1.4M nps = 2600 Elo > >2.8M nps = 2650 Elo > >5.6M nps = 2700 Elo > >11.2M nps = 2750 Elo (equal to Fritz 7) > >22.4M nps = 2800 Elo > >44.8M nps = 2850 Elo > >89.6M nps = 2900 Elo > >179.2M nps = 2950 Elo > > >Give me a tough problem that Fritz 7 can solve, and I will give you a Crafty at >200M nps that can solve it in less time. Hell, give me a problem that Fritz 7 >CANNOT solve, and I will give you a Crafty at 200M nps that can solve it. > >The arguement of "software is better now" is getting pretty old. NO IT IS NOT. >Sure, 20%. 30% maybe. Hardware has done more for computer chess in the last 7 >years, than will be done in the next 25 for software (something to the tune of >400%). Read: http://wondersmith.com/rants/howfar.htm. > >Junior 6 and Junior 7 are seperated by a mere 25 Elo on the SSDF, on the same >exact HW. However Junior 7 on an AMD 1200mhz is 84 Elo points stronger than >Junior 7 on an AMD K6/2 450mhz. I know that Deep Junior 7 gets around 2M nps on >my 2x1.53Ghz system. Now imagine raising that 200M. > >I attempted to run the Hyatt vs Ed "NPS Challenge", however due to circumstances >beyond my control, I had to stop after 4 40/120 games and 1 (adjourned) 40/120 >vs 40/12000 game. Rebel 4 won ALL 40/120 games, and was down 2.xx when I had to >adjourn the time-odds game. (Ed told me he was going to send me the real >version of Century 4, but it never made it, and my "trial time" expired, which >stopped me from using it anymore.) > >NPS is king. 15 years of HW advancements has given computer chess a 400% boost. > Another 15 years *STILL* won't give us DB NPS numbers.
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.