SSDF Computer Rating List 4/98 (09-JUN-1998)


60623 games played by 179 computers

                                           Rating   +     -  Games   Won  Oppo
                                           ------  ---   --- -----   ---  ----
   1 Fritz 5.0 PB34% 67MB P200 MMX           2573   27   -25   784   69%  2430
   2 Nimzo 98 52MB P200 MMX                  2537   32   -31   502   61%  2455
   3 Hiarcs 6.0 41MB P200 MMX                2536   31   -30   552   61%  2460
   4 Rebel 9.0 36MB P200 MMX                 2525   31   -30   550   67%  2403
   5 MChess Pro 7.1 35MB P200 MMX            2519   28   -28   630   60%  2450
   6 MChess Pro 6.0 38MB P200 MMX            2514   31   -30   540   60%  2445
   7 Rebel 8.0 32MB P200 MMX                 2510   43   -41   293   66%  2397
   8 Genius 5.0 DOS 35MB P200 MMX            2497   29   -28   614   60%  2428
   9 Shredder 2.0 51MB P200 MMX              2476   34   -34   429   51%  2468
  10 Hiarcs 6.0 Pentium 90 MHz               2448   22   -22  1015   56%  2405
  10 Rebel 9.0 Pentium 90 MHz                2448   29   -29   572   52%  2436
  12 Rebel 8.0 Pentium 90 MHz                2447   20   -20  1193   58%  2388
  13 Hiarcs 5.0 Pentium 90 MHz               2432   39   -38   343   61%  2355
  14 Genius 5.0 DOS Pentium 90 MHz           2428   22   -22  1044   53%  2410
  15 MChess Pro 6.0 Pentium 90 MHz           2421   20   -20  1230   52%  2409
  16 Rebel 6.0 Pentium 90 MHz                2408   31   -30   540   60%  2336
  17 Genius 3.0 Pentium 90 MHz               2407   27   -26   719   62%  2320
  17 MChess Pro 5.0 Pentium 90 MHz           2407   26   -25   759   62%  2318
  19 Genius 4.0 DOS Pentium 90 MHz           2406   24   -24   877   61%  2328
  20 Rebel 7.0 Pentium 90 MHz                2403   26   -25   749   61%  2322
  21 Junior 4.0 Pentium 90 MHz               2395   26   -27   692   46%  2420
  22 Hiarcs 4.0 Pentium 90 MHz               2394   25   -25   766   55%  2359
  23 Shredder 1.0 Pentium 90 MHz             2388   59   -58   145   53%  2369
  24 Chessmaster 5000 Pentium 90 MHz         2387   49   -45   240   67%  2262
  25 Nimzo 3.5 Pentium 90 MHz                2386   26   -26   721   49%  2392
  26 Nimzo 3.0  Pentium 90 MHz               2381   26   -25   767   58%  2323
  27 Hiarcs 3.0 Pentium 90 MHz               2375   30   -30   545   56%  2330
  28 R30 v. 2.5                              2372   44   -42   283   66%  2253
  29 Genius 4.0 DOS 486/50-66 MHz            2369   23   -23   880   50%  2369
  30 Junior 3.3-3.5  Pentium 90 MHz          2364   31   -31   496   48%  2379
  31 Genius 3.0 486/50-66 MHz                2357   24   -23   910   62%  2269
  32 MChess Pro 4.0 Pentium 90 MHz           2356   29   -29   578   52%  2340
  33 Fritz 3.0 Pentium 90 MHz                2347   21   -21  1075   48%  2364
  34 Mephisto Genius 2.0  486/50-66 MHz      2337   23   -23   917   58%  2277
  35 Fritz 4.0 Pentium 90 MHz                2335   40   -39   324   60%  2264
  36 WChess 1.06 Pentium 90 MHz              2330   25   -25   780   46%  2361
  36 MChess Pro 5.0 486/50-66 MHz            2330   26   -26   691   50%  2332
  38 Rebel 7.0 486/50-66 MHz                 2329   26   -26   737   50%  2331
  39 Kallisto 1.98 Pentium 90 MHz            2324   24   -24   870   43%  2372
  40 MChess Pro 4.0 486/50-66 MHz            2311   24   -24   846   53%  2287
  41 Chess Machine 30-32 MHz Schröder 3.1    2302   32   -30   546   68%  2170
  42 WChess 1.03 486/50-66 MHz               2300   27   -27   677   49%  2306
  43 Rebel 6.0 486/50-66 MHz                 2298   24   -23   887   55%  2264
  43 Hiarcs 3.0 486/50-66 MHz                2298   24   -24   865   50%  2295
  45 Meph Genius 68 030 33 MHz               2297   47   -46   228   57%  2251
  46 Ch.Machine 30 MHz King 2.0 aggr/R30 off 2294   21   -20  1202   64%  2193
  47 Chessmaster 4000 486/50-66 MHz          2292   34   -33   462   66%  2172
  48 Chess Genius 1.0  486/50-66 MHz         2283   23   -23   931   54%  2256
  49 Rebel Decade Pentium 90 MHz             2277   29   -30   563   45%  2315
  50 Chess Machine 30 MHz Schröder 3.0       2274   62   -55   168   73%  2102
  51 Mephisto Gideon Pro  486/50-66 MHz      2271   35   -34   421   62%  2183
  52 MChess Pro 3.5  486/50-66 MHz           2269   23   -22   958   51%  2260
  53 MChess Pro 3.12  486/50-66 MHz          2265   36   -34   438   69%  2122
  54 Fritz 3.0 486/50-66 MHz                 2259   20   -20  1206   46%  2287
  55 Mephisto Lyon  68030 36 MHz             2252   28   -27   716   68%  2117
  56 Chess Genius 1.0  486/33 MHz            2241   32   -32   463   50%  2241
  57 Mephisto Portorose  68030 36 MHz        2238   40   -36   407   77%  2028
  58 Mephisto Vancouver  68030 36 MHz        2227   35   -33   471   71%  2070
  59 MChess Pro 3.12  486/33 MHz             2226   50   -48   208   60%  2152
  60 Berlin Pro 68 020 24 MHz                2225   24   -24   850   58%  2170
  61 Kasparov SPARC  20 MHz                  2223   24   -24   845   48%  2235
  62 Comet32 Pentium 90 MHz                  2213   25   -27   830   26%  2399
  63 Kallisto 1.82-1.83 486/50-66 MHz        2212   21   -22  1077   41%  2274
  64 Mephisto RISC 1 MB ARM 2  14 MHz        2200   21   -20  1174   56%  2158
  64 Hiarcs Master 2.0  486/33 MHz           2200   46   -46   229   51%  2191
  66 MChess 1.1-1.71  486/33 MHz             2193   44   -40   326   74%  2014
  66 Saitek RISC 2500 ARM2 14 MHz 128K       2193   21   -21  1116   55%  2159
  66 Chess Machine Schröder 512K ARM2 16MHz  2193   27   -26   703   62%  2109
  69 Mephisto Montreux ARM  14 MHz 512K      2187   32   -30   589   73%  2015
  70 Chess Machine The King 512K ARM2 16MHz  2174   32   -32   473   55%  2136
  71 Mephisto Vancouver  68020 12 MHz        2159   23   -22  1001   66%  2046
  72 Mephisto Lyon  68020 12 MHz             2150   19   -18  1442   59%  2087
  73 Socrates 3.0  486/33 MHz                2140   49   -50   203   47%  2161
  74 Fritz 2.0  486/33 MHz                   2130   30   -31   527   45%  2164
  75 Mephisto Portorose  68020 12 MHz        2127   26   -24   896   73%  1956
  76 M Chess 1.1-1.71 386/25-33 MHz          2126   36   -35   408   65%  2017
  77 Sapphire II                             2125   43   -40   308   70%  1980
  78 Fidelity Elite  68030 32 MHz (vers.9)   2122   31   -30   521   60%  2049
  79 Mephisto Berlin  68 000 12 MHz          2116   23   -22   966   57%  2064
  80 Mephisto Lyon  68000 12 MHz             2104   21   -21  1130   58%  2050
  81 Mephisto Vancouver  68000 12 MHz        2103   22   -21  1062   56%  2062
  82 Mephisto MM 4 Turbo Kit  6502 16 MHz    2093   46   -42   292   73%  1918
  83 Novag Sapphire H8 10 MHz                2089   22   -22   994   48%  2107
  84 Milano Pro  SH7000 20 MHz               2086   39   -37   369   68%  1952
  85 Fidelity Mach IV  68020 20 MHz          2074   18   -18  1466   49%  2083
  86 Mephisto Almeria  68020 12 MHz          2073   42   -40   305   62%  1985
  87 Hiarcs Master 1.0  486/33 MHz           2070   48   -48   214   48%  2086
  88 Mephisto Portorose  68000 12 MHz        2045   25   -24   827   62%  1961
  89 Mephisto Polgar  6502 10 MHz            2042   43   -41   283   59%  1979
  90 Fritz 1.0  486/33 MHz                   2041   48   -47   215   55%  2008
  91 Gandalf 2.1 Pentium 90 MHz              2039   46   -52   236   27%  2213
  92 Fidelity Elite  68000 x 2 (vers. 5)     2037   49   -48   209   57%  1987
  93 Nimzo 2.2.1 486/33 MHz                  2033   46   -47   229   42%  2088
  94 Mephisto Roma  68020 14 MHz             2030   27   -26   712   65%  1919
  94 Rex Chess 2.3  386/25-33 MHz            2030   65   -62   126   59%  1964
  96 Zarkov 3.0 486/25-33 MHz                2029   46   -48   232   39%  2108
  97 Mephisto Dallas  68020 14 MHz           2023   35   -32   492   72%  1861
  98 Mephisto Almeria  68000 12 MHz          2021   32   -30   534   67%  1898
  99 Zarkov 2.5  386/25-33 MHz               2020   56   -53   168   61%  1941
 100 Kasparov Brute Force  H8 10 MHz         2018   23   -23   906   44%  2058
 101 Fritz 1.0  386/25-33 MHz                2007   66   -67   113   46%  2038
 102 Novag Diablo  68000 16 MHz              2005   21   -21  1123   41%  2071
 103 Fidelity Mach III  68000 16 MHz         1993   14   -14  2410   52%  1980
 104 Complete Chess System  486/33 MHz       1984   47   -47   221   47%  2006
 105 Kasparov President/GK-2100 H8 10 MHz    1979   29   -29   579   46%  2005
 106 Mephisto MM 5  6502 5 MHz               1975   19   -19  1345   47%  1993
 107 Mephisto Dallas  68000 12 MHz           1971   21   -20  1177   61%  1894
 108 Mephisto Polgar  6502 5 MHz             1970   17   -17  1773   41%  2035
 108 Mephisto Roma  68000 12 MHz             1970   18   -18  1519   56%  1927
 110 Mephisto Milano  6502  5 MHz            1964   24   -24   860   41%  2028
 111 Zarkov 2.6 386/25-33 MHz                1960   64   -62   125   56%  1917
 111 Novag Super Expert C  6502 6 MHz Sel 5  1960   18   -18  1545   46%  1990
 113 Mephisto Academy  6502 5 MHz            1939   18   -18  1535   44%  1982
 114 Chessmaster 3000  386/25-33 MHz         1932   68   -74   103   36%  2031
 115 Rex Chess 2.3  386/16-20 MHz            1930   53   -53   174   49%  1936
 116 Saitek Leonardo Maestro B 6502 18 MHz   1928   73   -67   105   62%  1842
 117 Mephisto Amsterdam  68000 12 MHz        1927   22   -22  1020   58%  1873
 118 Novag Super Expert B 6502 6 MHz sel 4   1919   29   -29   564   48%  1935
 119 Mephisto Mega IV  6502 4.9 MHz          1916   17   -18  1584   44%  1960
 119 Saitek Maestro D  6502 10 MHz           1916   26   -26   736   51%  1909
 121 Fidelity Excel Mach IIc  68000 12 MHz   1915   23   -23   955   50%  1911
 122 Novag Expert Turbo Kit 6502 16 MHz      1910   67   -63   120   61%  1829
 123 Mephisto MM 4  6502 5 MHz               1904   19   -19  1288   51%  1900
 124 Fidelity Travel Master H8/330 10 MHz    1897   65   -79   123   22%  2112
 124 Mephisto Modena  6502 4 MHz             1897   27   -28   655   40%  1968
 126 Kasparov GK-2000   H8 10 MHz            1896   29   -29   593   42%  1952
 127 Psion Atari  68000 8 MHz                1882   18   -18  1487   44%  1928
 128 CXG Sphinx Galaxy  6502 4 MHz           1880   17   -17  1710   37%  1974
 129 Novag Ruby H8  10 MHz                   1879   30   -30   545   42%  1934
 130 Fidelity Excel Mach IIa  68000 12 MHz   1878   46   -47   226   47%  1898
 131 Saitek Turboking II  6502 5 MHz         1868   23   -23   963   37%  1964
 132 Conchess Plymate Victoria 6502 5.5 MHz  1867   26   -27   701   40%  1941
 133 Fidelity Excel Club  68000 12 MHz       1857   23   -23   931   51%  1849
 134 Mephisto Mega IV Brute Force            1847   35   -36   402   38%  1930
 135 Novag Super Expert/Forte A 6502 5 MHz   1837   19   -19  1358   38%  1923
 136 Fidelity Par Excellence  6502 5 MHz     1835   22   -22  1022   48%  1846
 136 Fidelity Avant Garde  6502 5 MHz        1835   20   -20  1196   46%  1866
 138 Mephisto Rebell  6502 5 MHz             1824   19   -19  1393   38%  1909
 139 Saitek Stratos/Analys B  6502 6 MHz     1820   20   -21  1213   38%  1908
 140 Mephisto Super Mondial  6502 4 MHz      1819   19   -19  1375   37%  1914
 141 Novag Forte B  6502 5 MHz               1816   24   -24   837   44%  1860
 142 Saitek Leonardo Maestro A  6502 6 MHz   1811   30   -30   547   45%  1846
 143 Novag Forte A  6502 5 MHz               1809   22   -22   964   48%  1822
 144 Conchess Plymate  6502 5.5 MHz          1808   15   -16  2040   40%  1881
 145 Saitek Simultano/Corona C  6502 5 MHz   1806   23   -24   969   33%  1932
 146 Fidelity Excellence  6502 4 MHz         1801   18   -18  1565   43%  1852
 147 Novag Expert  6502 4 MHz                1788   24   -24   843   42%  1847
 148 Fidelity Elegance  6502 3.6 MHz         1781   40   -40   303   54%  1752
 149 Conchess Plymate  6502 4 MHz            1780   39   -38   333   55%  1748
 150 Mephisto MM 2  6502 3.7 MHz             1773   49   -48   208   52%  1756
 151 Saitek Turbostar 432  6502 4 MHz        1765   22   -22  1005   45%  1802
 152 Fidelity Excellence  6502 3 MHz         1757   22   -23   960   42%  1815
 153 Saitek Kasparov Blitz                   1734   45   -47   239   37%  1825
 153 Novag Super Nova                        1734   37   -38   350   40%  1803
 155 Novag Super Constellation  6502 4 MHz   1731   18   -18  1626   35%  1840
 156 Conchess Glasgow  6502 4 MHz            1720   39   -39   327   50%  1722
 157 Chessplayer 2150 Atari/Amiga            1714   58   -63   146   33%  1837
 158 The Final Chesscard  6502 5 MHz         1700   57   -64   156   27%  1872
 159 Novag Super VIP                         1688   53   -58   174   32%  1817
 160 Chessmaster 2100 Amiga 68000            1687   65   -76   120   25%  1874
 161 Mephisto Europa                         1683   50   -52   190   42%  1741
 162 Saitek Superstar 36K  6502 2 MHz        1679   23   -24   954   31%  1819
 163 Fidelity Elite A/S  6502 3.2 MHz        1678   22   -23  1043   28%  1842
 164 Chess Champion 2175 Atari/Amiga         1673   56   -62   157   30%  1822
 165 Conchess Glasgow  6502 2 MHz            1667   24   -25   855   34%  1782
 166 Novag Quattro  6502 4 MHz               1663   30   -31   560   33%  1790
 167 Fidelity Prestige  6502 4 MHz           1661   64   -66   118   44%  1703
 168 Chessmaster 2000 Atari  68000 8 MHz     1654   41   -45   312   25%  1841
 169 Novag Constellation  6502 3.6 MHz       1648   31   -32   504   41%  1715
 170 Novag Constellation Primo               1638   57   -59   149   44%  1683
 171 Novag Constellation  6502 2 MHz         1597   32   -33   479   35%  1706
 172 CXG Super Enterprise                    1563   36   -39   386   28%  1727


Rating : ELO rating
+/-    : deviation
Games  : number of games played
Won    : percentage of games won
Oppo   : medium opponent ELO rating


Comments to the Ratinglist 4/98 (09-JUN-1998)

By Thoralf Karlsson, thoralf.karlsson@mailbox.swipnet.se, Chairman of the SSDF

Unfortunately we don't have any new programs or chess computers on this list. But we have played 1.261 games in order to get more reliable ratings for the available programs. Fritz 5.0 P200 MMX is still number one on the list, but it's rating has gone down 16 points. The difference between the first and second program has diminished from 55 to 36 points.

Several of the programs in the upper part of the list have lost a few points. Shredder 2.0 has suffered the biggest loss with 24 points down. Sapphire II has gone up 15 points, and is now 36 points higher than it's predecessor.

As can be seen, the rating list contains some new information. For the programs tested on P200 MMX, the average amount of used RAM is shown. It can be seen that some of the older programs have had slightly more than 32 MB RAM, that a couple of programs have used about 50 MB and that Fritz 5.0 has had little more than 64 MB RAM in average. More about that further down.

For Fritz 5.0 you can see in how many percent of the games the opening book PowerBook has been used. This opening book is sold separately. In 520 of the 784 games the ordinary book fritz5.ctg has been active, and PowerBook has been used in the other 264 games. If you make a list with only the 520 games with the standard opening book, Fritz 5.0 gets 2561. With only the 264 PowerBook-games, Fritz 5.0 would receive 2599. The difference is 38 points. This could be interpreted to show that PowerBook raises the playing strength. But the margin of error is 30 and 45 points respectively. Further, the games have been played against different opponents, which makes a comparison more difficult.

At the moment we are playing with CometA90 P200 MMX and Mephisto Atlanta. In a short time we will start to test Kallisto II and later we will play with Junior 5.0 and probably also with CST. A week ago we recived the danish program Gandalf 3. If time permits we will try to establish it's playing strength against other programs. Later this year we have been promised to receive Hiarcs 7.0, MChess Pro 8.0 and a new Nimzo-version.

Next rating list is expected to be released in early August.


SSDF and Fritz 5.0

Since the arrival of the rating list dated 98-02-22, where Fritz 5.0 unexpectedly received a rating 55 points ahead of it's competitors, SSDF has been the target of heavy attacks from various sources. The critics have in many cases been connected with some of the other strong programs.

The essence of the criticism has been that SSDF has given Fritz 5.0 an unfair advantage in various ways, and that this program isn't worth it's high rating. Although many of the arguments and allegations have disappeared from the debate now, I would like to comment on the most important issues concerning Fritz 5.0 and it's rating.

Fritz 5.0-versions
It was claimed that SSDF didn't play with a commercially available version of the program. That's not true. The chess engine which we have tested has been for sale since November 97, and has sometimes been called the Paris version. Owners of the original Fritz 5 version (sold summer 97 until October 97) can purchase an upgrade from ChessBase. SSDF does not know if there is any difference in playing strength between the original version and the one which has been for sale since November 97.

Our games are published on Tony Hedlunds website (new address: http://home.interact.se/~w100107/welcome.htm ) and several persons have verified that they can reproduce the played games with their own Fritz 5 versions.

PowerBook
Chess programs often have different opening books, for example a more narrow tournament book and a wider and larger book with more variety. SSDF has had the policy to use the opening book, which the programmer recommends. For Fritz 5.0 ChessBase recommended the opening book named PowerBook. It has been said that this opening book isn't for sale, but that's not true. It is sold separately.

It was our intention to use PowerBook consistently. Due to small hard disks, misunderstandings and failings to load the PowerBook, it was only used in 18% of the games until February 22nd and in 34% of the games until now. Some persons have the opinion that we should only use opening books which are delivered together with the program. For us the important thing is that the used opening book is for sale and that it is evident from the list or the comments which book we have used. If our testing capacity were larger it would be a good idea to test programs with different opening books, but that's normally not possible.

64 MB RAM
It has been said that Fritz 5.0 had access to 64 MB RAM whereas the other programs only had 16 or 32 MB RAM. This is partly true and partly untrue. First, no program tested on P200 MMX has had only 16 MB. When we started to use this hardware for more than a year ago, we decided to use at least 32 MB RAM. We have never had an upper limit for the amount of RAM. If one of the testers bought more RAM because he needed it, this didn't make us disqualify his games. Since a doubling of the hashtables has increased the speed about 6-7%, something which could give 4-5 additional rating points, we haven't bothered about it. Programs tested early in a "hardware cycle" (486/33, 486/66, P90) has always in average had less RAM memory than those tested later on.

There are practical and economical reasons for this. Either you should try to force all testers to use for example 4 or 8 MB RAM during 2-3 years, or you should demand that all testers in spite of very high costs bought a very large RAM memory from the beginning. Neither alternative has seemed to be reasonable with the RAM prices during the nineties. (Note that SSDF doesn't have money to sponsor the hardware which the tester owns and uses.)

64 MB RAM was first used during the summer of 97. Hiarcs 6.0 was the first program to have this possible benefit. Until now all programs on P200 MMX except Rebel 8.0 partially has had access to 64 MB RAM. The two programs which we tested just before Fritz 5.0 had had more than 32 MB of RAM (mostly 64 MB) in more than 60% of the games in the list from February 22nd.

Thus,when we started to play automatically with Fritz 5.0, we were in a process of upgrading our RAM memories from 32 to 64 MB RAM. In this situation I judged it to be acceptable that Fritz 5.0 was tested solely with 64 MB RAM. Normally we should have tested it with the mixture of different RAM sizes which were available at that time, but the autoplayer only worked with 64 MB RAM. We could of course have demanded that the autoplayer should be reprogrammed, but since in our opinion it was only a question of a few points, it wasn't worth delaying the testing because of this. By the way, upgrading from 32 to 64 MB RAM costs only about 40 US dollars.

I have asked the programmers about the speed difference between 32 and 64 MB RAM for their programs. None of them has claimed that it would be be a question of more than a few points, and our own tests supports that. It might be possible that Fritz 5.0 can benefit more from larger hash tables than other programs. But the figures differs. One investigation indicates that a doubling would in average give a speed increase of 18%, whereas another shows the usual 6-7%. But if your comparison is made with 64 MB RAM as the starting point, the disadvantage for the other programs would not be larger than 1-4 points.

It has been claimed that SSDF received hardware from ChessBase. That is not correct. We were offered to borrow a machine with 128 MB RAM, but declined this offering since only Fritz 5.0 and no other programs would have had the possibility to be tested with this machine.

In this table you can see which RAM sizes the different programs have used until the latest rating list from 09-JUN-1998:

  96 MB 64 MB 48 MB 32 MB
Rebel 8.0 0% 0,0% 0,0% 100,0%
MChess Pro 7.1 0,0% 6,3% 3,2% 90,5%
Genius 5.0 0,0% 7,8% 3,6% 88,6%
Rebel 9.0 0,0% 10,2% 7,3% 82,5%
MChess Pro 6.0 0,0% 15,6% 0,0% 84,4%
Hiarcs 6.0 0,0% 26,8% 3,6% 69,6%
Shredder 2.0 4,7% 43,8% 14,0% 37,5%
Nimzo 98 0,0% 59,6% 3,9% 36,5%
Fritz 5.0 10,2% 89,8% 0,0% 0,0%

With 48 MB RAM Shredder 2.0 can have the same size on it's hash- and evaluationtables as is recommended for 64 MB RAM. So it appears that 48 MB here is as good as 64 MB. It's doubtful whether 96 MB really has been an advantage for the two programs which partially have used it. For the commonly used motherboards for P200 MMX it seems that the cache-memory cannot handle more than 64 MB RAM. More RAM than 64 MB could even slow things down. 96 MB RAM is not used any longer.

Since several months all test machines have 64 MB RAM. This, and only this, size will be used together with P200 MMX in the future. All new programs will be tested only with 64 MB RAM. As we continue to play more games also with the older programs, the average RAM size will slowly come closer to 64 MB RAM.

Problems may arise when it's time to change the hardware. If we for example next year start to use Pentium II 450 MHz, should we then choose 64, 128 or perhaps 256 MB? 64 MB would probably suit the economy of the testers best, but it might seem much too small two years later. If we decide to make 256 MB RAM the standard, it might have the effect that most testers refrain from buying a new PC. Or should we choose 64 MB RAM as minimum and let the tester buy more RAM when prices eventually go down? Having the same RAM size during the whole hardware-cycle is of course ideal, and if possible we will try to have it so. But we don't make any binding decisions about that today.

Autoplayer cheating?
For some time it was claimed that Fritz 5's autoplayer cheated, intentionally or unintentionally. It seems clear that at least is possible to do some kind of cheating, for example by forcing the opponent to make it's move immediately. But has it been done? Ed Schröder made a special version of Rebel 9, in which a logfile recorded the commands which were sent between the computers. Enrique Irazoqui then tested Fritz 5.0 and other programs against this Rebel version, using the AUTO232 protocol. Their conclusion was that nothing strange happened and that the autoplayer was "clean"! Thereby the hypothesis of cheating can be dismissed.

Not generally available autoplayer
The autoplayer which SSDF has used for testing Fritz 5.0, is not sold together with the program and cannot be purchased separately. The reason for this seems to be that ChessBase want to make it more difficult for the competing programmers to make opening book preparations against Fritz. (ChessBase claims that Fritz 5.0 in it's opening books uses only moves played in human tournaments or known theoretical lines, and that no special preparations have been made against the other chess programs.)

It has been said that it is unfair that ChessBase can play automatically against the other programs but that the other programmers cannot do the same against Fritz 5.0. It's still possible to play manual games, but you need someone who transfers the moves, and he might want to be payed for it. Practically speaking it will also take longer time, since you'll probably not find someone willing to do this work during nights.

So if playing games against the competitors is part of the work with a new program (and that is probably the case for most programmers), the production cost will be higher if you cannot play automatically against Fritz 5.0. I don't know against how many other programs the chess programmers play, but I guess against 5 - 10. If so, 10-20% of the games cost more to play.

The primary goal for the SSDF rating list is to provide information about the relative strength of chess programs. Whether the production costs have been somewhat higher or lower is no primary concern for us.

We understand that the rating list can be looked upon as a kind of world championship for chess programs, and in such an event the conditions should be fair and equal. The question is how to define fairness. Equal hardware, if possible the same RAM size, roughly the same amount of games against the same opponents are fair demands.

But should the conditions for making the chess program be regulated by SSDF? Some programmers work full time, others make their program during evenings and weekends. Is that fair? Some are working alone, others have several employees who help them. Is that fair? Some have better economy and can afford more PCs for automatic play and stronger chess players as advisors. Should SSDF make rules here?

So far I haven't found strong enough theoretical arguments for a rule stating that SSDF in the future would only test programs which have generally available autoplayers. We have discussed the possibility of such a rule just for pragmatical reasons. Our wish has always been to include all of the strongest programs in our list. But as it seems now this will not be the case in the future, no matter what decision we make. Whether one of the alternatives would give us more programs to play with than the other is an open question. (Either SSDF would refuse to play with programs without autoplayers, like Fritz 6.0(?), Virtual Chess, CM6000(?) and all chess computers, or in the other case some programmers would disable the autoplayers in their coming programs in order to make it impossible or very difficult for SSDF to test them.)

The ideal solution for SSDF would have been that ChessBase released their autoplayer for Fritz 5.0, something which they promised to do (under certain conditions). I have argued with ChessBase that these conditions are fulfilled, but as it seems they will not release the autoplayer for Fritz 5.0. They will however release the autoplayer for the coming engines Junior 5.0, Nimzo and Crafty. Nothing was said about Fritz 6.0.

Concluding remarks
Much more could be said, but then the rating list would be even more delayed! In our swedish magazine PLY more details are discussed. There the members of SSDF are invited to present their opinions.

For the time being we continue as before, with the exception that we now inform about RAM sizes and usage of separately sold opening books. And we will use only 64 MB RAM in the future, for programs tested on P200 MMX.

As it seems now we will continue with the rating list as long as we have willing testers and programs to play with. If the chess programmers make it impossible for us to play with their chess programs, then we have to choose another hobby.

Finally I would like to thank all those who have defended SSDF during the last months!! And I would also like to thank those who have sent emails expressing their support and their wish for the continued publication of the rating list!




Last updated: Wednesay, 1. July 1998
Copyright Manfred Rosenboom marochess@oocities.com


This page hosted by Get your own Free Home Page