Author: Per Jørgensen
Date: 18:12:24 11/17/05
Go up one level in this thread
On November 17, 2005 at 15:11:15, Ómar Skúlason wrote: >On November 17, 2005 at 13:06:05, Per Jørgensen wrote: > >>Hello Chesscomputerfriends! >> >>Several Months ago I got the idea to make a Ratinglist similar to the one Klaus >>Wlotzka is making for CSS (Computer Schach und Spiele) but with one important >>difference. While Klaus is testing with 10 min. for the whole game plus 10 sec. >>per made move I would be testing with only half the time (5+5). Fortunately >>Klaus thought it was a good idea and I would like to thank him for support and >>advice during the process. I did a lot of testing and I managed to develop a >>Ratinglist of 17 Engines (each one having played 320 games). At that time I >>realized that my Hardware wasn't strong enough to cope with the needs of modern >>Engines (especially Fritz 9 which in my opinion requires at least 512 kb Level 2 >>Cache to perform at its maximum). Consequently I have purchased new Hardware: >> >>Asus A8V Motherboard, Socket 939, FSB 1.000 MHz >>AMD Athlon 64 4000+, 2.4 GHz, 1 MB Level 2 Cache, 90 nm San Diego Core >>2 x 512 MB Corsair PC 3200XLPT RAM (CAS Latency 2-2-2-5) >> >>A couple of days ago I started the new testseries and I'm still playing with the >>thinking time 5+5. As mentioned earlier I am testing the way Klaus Wlotzka is >>doing: 10 basic openingpositions that each engine has got to play with white and >>black meaning that an Enginematch equalizes 20 games. If you are interested in >>further details (like the specific openingpositions) I refer to the excellent >>Website of CSS: www.computerschach.de. It's my aim, just like Klaus Wlotzka, to >>develop a Ratinglist of app. 25 Engines. So far I have been testing 5 Engines >>and here are the results: >> >>AMD Athlon 64 4000+ >>64 MB PC 3200 Ram (CAS Latency 2-2-2-5) >>8 MB Cache for TB's >>TB's: Nalimov - 3 + 4 Pieces >>Pondern Off >> >>Fritz 9 - Shredder 9: 11½-8½ >>Spike 1.0a Mainz - Fritz 9: 7½-12½ >>Spike 1.0a Mainz - Shredder 9: 8½-11½ >>Fruit 2.2.1 - Fritz 9: 6-14(!) >>Fruit 2.2.1 - Shredder 9: 11½-8½ >>Fruit 2.2.1 - Spike 1.0a Mainz: 14-6 >>The King 3.33 - Fritz 9: 7½-12½ >>The King 3.33 - Fruit 2.2.1: 6-14 >>The King 3.33 - Shredder 9: 9½-10½(!) >>The King 3.33 - Spike 1.0a Mainz: 10-10 >> >>And here is the current standing in points after each engine have played 80 >>games: >> >>1. Fritz 9 50½ >>2. Fruit 2.2.1 45½ >>3. Shredder 9 39 >>4. The King 3.33 33 >>5. Spike 1.0a Mainz 32 >> >>I guess that this standing is not a big surprise but at least 2 things are >>remarkable: 1) The King 3.33 have so far made a very strong performance and >>perhaps this huge and comprehensive Programme benefits from my quite fast >>Hardware and 2) Although each engine have only played 80 games Fritz 9 leads >>Shredder 9 by more than 10 points! When I made my last Ratinglist on my old >>Hardware Shredder 9 was app. 30 Ratingpoints ahead of Fritz 9, I certainly don't >>expect this to happend again! >> >>The next engines to be tested are Ruffian 2.1.0 and Loop List. >> >>Best regards >>Per Jørgensen > >Hi Per! > >The report of your fine hardware update; got me to think >about what is an _ordinary_ hardware for us "hometesters". >We sometimes talk about "the strongest engine on ordinary hardware" or >"strongest engine for 1 cpu". >I thougt for a moment that you were favoring Fritz with "512 level 2 cache" >but I guess it is still ordinary, at least it was in my comp when I >looked just now. >But I think that you must get you all 5 men TB if you want to be >respected :-) >And.. Ruffian is old news, I would rather see you test GambitFruit with >every parameter turned on including the mistic Rebel Reduction. > >Best regards, Ómar. Hi Omar! Thanks for your comments. Regarding your reflections on Hardware, I'd say that there ain't a typical "ordinary" Hardware for all us "hometesters". I'd guess that most of prefer to have a very fast hardware (at least it might be a problem if you are testing on a very slow Hardware because these results don't have the same authority or respect as tests made on a fast Hardware). I wouldn't say that I am favoring Fritz 9 but I discovered that Fritz 9 wasn't performing as well as it "should" (compared to the majority of tests I've seen on the Internet). A wise man told me that if the size of the Level 2 Cache is below 512 kb, well then Fritz 9 cannot perform at its maximum. I'd guess he's right if you take a look at my new results. But basicly every engine benefits from my new Hardware; instead of a Sempron 2.0 GHz CPU with 256 kb Level 2 Cache I am now testing on a AMD Athlon 64 CPU 2.4 GHz with 1 MB Level 2 Cache. This is ofcourse Great News for each Engine but it is true that some engines (like Fritz 9) benefits more than others from the faster Hardware. Let me give you a small example to illustrate what has happened: These followings figures (positions analyzed per second) are observed in a random middlegameposition on my old and new Hardware for Shredder 9 and Fritz 9: Shreddder 9: app. 375-380.000 --> app. 570.000 (Factor 1,51) Fritz 9 : app. 900.000 --> app. 1.442.000 (Factor 1,60) Quite a difference and as you see it seems as Fritz 9 benefits most of these two engines and the results so far confirm this. Regarding TB's we all know that it's a subject where it's hard to find consensus. According to some testers large TB's are very important while other tend to believe that TB's aren't an advantage at all (and that you should only use this tool for analyzing matters). Well, I'd say that at least theoretically it would be an advantage if I was testing with all 5 men TB's but I'd guess that when it comes down to the endrating of each engine there ain't a big difference (maximum 5-10 ratingpoints) whether you are doing your tests with 4 or 5 men TB's. In due time I might test Gambit Fruit but right now this is only a Betaversion and I am only testing "official" versions. Ofcourse Ruffian 2.1.0 is not the newest engine on the market but when you want to create a ratinglist of app. 25 engines (the strongest on the market) you have to include engines like Ruffian 2.1.0 as well. Best regards Per :)
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.