My 'average club player, class B level' chess computers

Fidelity Sensory Chess Challenger 9 'b'

Year: 1983
Programmer: Dan & Kathe Spracklen
CPU: 6502 @1.88Mhz
ROM: 16Kb
Elo level: 1610 (1599 FIDE)
CMhz: 1.88
Rperf: 93%
Square size: 1"

This electronic chessboard also left a mark on the chess computers market, offering a strong play, close to high-end ones at that time, at affordable cost. For a while I assumed it too close to the Excellence to look after one, but considering the later one too tough for me, the Sensory 9 was getting somewhat attractive, and I owned no chess computer featuring a LED per each square, this was another attractive point for buying. What did finally trigger my decision? This vote establishing a nice 12th ranking for the Sensory 9 among most desirable chess computers. I am happy with this buy, considering the '9' is actually a tough player, but I can still beat it, and it is pleasant to use. Its design is less dreary than the Excellence one. I was a bit lucky to discover the '9' I received (bought €40 on French 'le bon coin') was actually a '9b' (recognizable through its serial number 32xxxx and more surely through its answer 1. ...... b6 to 1. h4). The '9b' features a slightly evolved software (it is reknown to benefit from some enhancements brought with the 'Budapest' program hosted in the Elite Auto Sensory) and a higher speed clock than first series running at 1.6Mhz. A '9b' is usually clocked with 2Mhz... However this device seems to belong to very early '9b' series; its production batch date is 1983, August 26th according to its serial number, its model number is incorrect (SU9 instead of SC9); and my accurate response time tests, later refined using an emulator, provide evidence its clock is actually 1.88Mhz.

Novag Constellation 3.6/Quattro

Year: 1986
Programmer: Dave Kittinger
CPU: 6502 @3.6Mhz
ROM: 16Kb
Elo level: 1642 (1624 FIDE)
CMhz: 3.6
Rperf: 91%
Square size: 0.98"

Novag reacts to keep a place in the race for strong chess computers at affordable price, in particular to counter the Sensory 9 success. The Constellation is initially offered with a 2Mhz clock in 1983, then upgraded to 3.6Mhz in 1984, and the most advanced version will be the famous 56K Super Constellation, 4Mhz, in 1984; a chess computer with a very high quotation. I point out the 'Super' mention in Novag range of products indicates a device with available connection for add-ons. I bought this '3.6' at a reasonable price (€60) to represent within my collection the 'good old 16K Kittinger' pre-2003, when the master still collaborated with Novag. Unlike the Carnelian II, only 8 levels are available, on the other hand permanent brain is active, and CPU is more than four times as powerful. I point out the undocumented ability to display the search depth (this 1986 example appears to run a Quattro ROM), following the same process as with a Super Connie: next to Set Level, push the bishop key (labelled Sound & Depth Search on the Super Connie) before completing the level setting using Go (watch video). The depth is displayed using base two numbering with the  LEDs 8 (standing for 2^0, thus 1) to 5 (standing for 2^3, thus 8).
Update: another one with the very same ROM reference and produced during the same timeframe, reverse-engineered in order to emulate the 3.6, shows evidence it is not a modified device, but actually a Novag produced serial. The romdump is same as a Quattro ROM's one. One can assume Novag had to sell off a number of 3.6 casings and printed circuit boards, while the new ROM was already produced for the Quattro; with regards to compatibility, continuing to simultaneously produce the former ROM would have been an economical nonsense; as well reprinting the 3.6 casing and user's manual. And last but not least, from a marketing standpoint, announcing a new 3.6 being a Quattro competitor would not have made much sense. Additional features brought by the Quattro ROM are: the already described depth search display, 8 training levels (fixed plies levels), and mate announcement (using 1-8 diods).
The design is rather pleasant, especially the pieces showing a moiré aspect:


CXG Super Enterprise

Year: 1986
Programmer: Kaare Danielsen
CPU: 6301Y @8/4Mhz
ROM: 16Kb
Elo level: 1673 (1647 FIDE)
CMhz: 1.68
Rperf: 97%
KT: 1260
Square size: 1.45"

Kaare Danielsen created a sensation the same year 1986, fighting mainframes and experimental chess computers in the WCCC (World Computer Chess Championship) in Cologne with his 16K program hosted in a portable Advanced Star Chess device. He will score 1.5 points out of 5, after beating an Amdahl 5860 and drawing a CDC Cyber 175, with his tiny 6301Y microcontroller! The 16K version is quite more difficult to find than the 4K one, despite having been used in several devices (thus CXG Super Entreprise and Advanced Star Chess, but Super Crown and Sphinx Titan as well; and also the Mephisto Merlin 16K). I watched after small ads for a long time, saw some poor condition ones, and finally bought for €38 this fine Super Enterprise featuring a large push sensitive board. I point out two additional queens are offered with this chessboard, a black one and a white one, this is not quite usual. The lengthy design of the pieces, all the same diameter at bottom, is exclusive; they are heavy weighted and pleasant to handle. Several playing styles can be chosen ("aggressive" being the most efficient one, according to Spacious_Mind's test); playing level is overall strong but a bit inconsistent (blend of strong and weaker moves).


Khmelnitsky test: the inconsistensy is patent, especially looking at the drop in playing level that happens between the opening phase, which is more than mastered, and the definitely weak endgame; with a well established middlegame in between. I point out I performed the test leveraging the aggressive style. The outcome is far from brilliant, the Super Enterprise achieved close to 80 KT points less than the Mirage Mephisto modul, ranked into the preceeding category! The stronger Elo of the Super Enterprise relies on the tactical domain (+177 KT points, compared to the Mirage) and the strategy (+138 KT points). These generic skills seem to have a deciding impact on a chess program performance! As most chess computers, the Super Enterprise is more comfortable counterattacking, rather than attacking; and this is very clear despite the choice I made to use the aggressive style.

Mephisto Europa A

Year: 1990
Programmer: Frans Morsch
CPU: 6301Y @8/4Mhz
ROM: 16Kb
Elo level: 1703 (1669 FIDE)
CMhz: 1.68
Rperf: 98%
Square size: 0.86"

It is the strongest chess computer powered by a 6301Y microcontroller (with standard speed). Its program derived from Nona authored by Frans Morsch does wonders with a relative performance similar to the Excellence one. It is particularly strong in tactics, a bit less in strategy, and reveals some weaknesses in the endgame. This small initiation chessboard has been largely sold in Germany, however I found mine in France for €40. Push sensitive squares are a bit tough, control panel keys are definitely hard, and the overall device size is a bit small, but its low cost and strength easily balance the drawbacks.



Novag Beluga

Year: 1990
Programmer: Dave Kittinger
CPU: 6301Y @16/4Mhz
ROM: 16Kb
Elo level: 1704 (1670 FIDE)
CMhz: 3.36
Rperf: 95%
KT: 1474
Square size: 0.98"

Please be aware chessmen shown here are not genuine parts: I bought this computer in perfect condition, in original packaging and with user's manual, but I received substitute chessmen, far too large and... ugly looking. This Beluga was nevertheless advertised on "le bon coin" French website with a snapshot showing original chessmen... I found too late evidence this snapshot had been copied from internet. It is good to know Google Images lets you search an image you can upload or locate using its URL; a precautionary measure I will now practice before buying. Anyway I was looking for a Beluga mainly for its play and for its cool chessboard look, and for €35 it would be ungracious to complain. I equipped it with magnetic wooden chessmen from a 10x10in travel chess set, they look like the ones from the Novag Carnelian 2 and fit nicely to my opinion. The Beluga is a direct competitor to the Mephisto Europa A: launched same year, sized and arranged the same way, it provides a couple of plus: the chessboard makes better use of available surface, thus being larger, moves and thinking information are displayed on the LCD screen (in addition to leds on coordinates), and the microcontroller is souped-up to double speed compared to most 6301Y, that is to say 4Mhz. A flaw to point out: the lack of rigidity of the chessboard, certainly caused by the too flexible plastic used and unsufficient support underneath. Despite the squares being sensitive and not requiring much pressure, the most central squares slightly bend the chessboard when pressed, just enough for a low-quality feeling.
Update: I purchased a Novag Super Nova (elder brother of the Beluga, using the same processor in the same casing, but with a 32K program) which revealed to be defective, but at least I could re-use its chessmen to restore the Beluga in its "original" shape:




Thanks to its display, the Beluga can run the Khmelnitsky test. Doing so, I noticed an interesting feature, as far as I know rather uncommon: as soon as a new position is entered in setup mode, you can request the score from either white or black point of view: the program responds using less than a second, and displays a static score (or very low depth score), mainly based on the present material (but not only: white and black scores are not symetrical, the side on the move has an influence on the absolute value). This enables a light check of the correctness of the input position, e.g. a balanced position should be scored close to zero. Fortunately, the same "Info" feature requested after an analysis by the computer provides the full-depth score. The skills profile is rather close to the ones of the Excellence and the Challenge, introduced hereafter, thus we may assume a mainly brute force approach, not using much selectivity. The Beluga gets rather well out of the opening phase, keeps up its position during middle game, before appreciably weakening in the endgame; where standard positions knowledge is bitterly lacking. Despite the relatively fast clock for a 6301Y, the microcontroller does no wonder for calculations. The Beluga is less comfortable attacking than counterattacking, but this small unbalance, usual amongst our chess computers from the same era, is not too much pronounced. Recognizing threats and tactics are, again as demonstrated by many others, sterling qualities. The nice surprise comes from the strategy domain, with a 1834 Elo points evaluation, the highest score in this category. In this dimension, the Beluga even overwhelms a number of 32K devices!

Tandy Chess Champion 2150

Year: 1988
Programmer: Julio Kaplan
CPU: 65C02 @3Mhz
ROM: 64Kb
Elo level: 1746 (1702 FIDE)
CMhz: 3
Rperf: 98%
KT: 1374
Square size: 0.99"

The interesting playing style of the Scisys Companion III made me want to get the 'heavy' version (actually a 'B' one) of Julio Kaplan's program (as always supported by Craig Barnes). 72K announced size for memory (including 8K RAM), a 3 Mhz 6502, a 100,000 half-moves opening book... It is the big brother of the former one, equipped with an advanced display able to show a number of data (preferred move, search depth, position evaluation) in addition to displaying the full chessboard. I pondered a while whether to buy the Scisys Turbo King, featuring the same (or close to) program, but the quotation is a lot higher for this one (I bought the CC2150 €60), the speed is still faster using a 5Mhz 6502 (too strong for me...) and I was keen to enter a Tandy (Radio-Shack) device in my collection, with a nostalgic regard to my old TRS-80.


Regarding seven out of twelve domains, the Tandy Chess Champion 2150 follows on the heels of the Fidelity Excellence, even outdoing it with calculations and tactics. On another hand, it is significantly weaker with strategy, standard endgame positions, and ability to recognize threats; despite this last domain being still at a good level. It engages its games fairly well, but gets weaker afterwards, and needs to hope for tactical opportunities in order to counterattack. It is of course relevant to compare it to the Saitek Turbo King II that runs a close program, but appreciably faster (5Mhz 6502 instead of 3 for the CC 2150): one can notice the fact that the weak points (strategy, endgame, attack) almost do not improve despite the speeding up, thus are intrinsic software weaknesses. The best enhancement thanks to the speed gain is the calculations ability, which is definitely logical (even if another cause is expected to help: a probably lighter evaluation code, please check the Turbo King II analysis next page).

Saitek Blitz

Year: 1990
Programmer: Julio Kaplan
CPU: H8 @10Mhz
ROM: 32Kb
Elo level: 1778 (1725 FIDE)
CMhz: 6.5
Rperf: 96%
KT: 1441
Square size: 1.23"

When I purchased the Blitz (priced €100 on 'le bon coin' including postage, so €86.25 for the chess computer, offered complete with its well preserved box), I already owned four chess computers from Julio Kaplan, and eight devices powered by a H8 processor... But I had no Kaplan running on H8! This one is the only of his programs running on this micro-controller. In addition to the Blitz, this program will be used in the Saitek Prisma (as well produced in 1990) and in the RadioShack Chess Champion 2150L (2 years later, using a 8Mhz clocking for the H8, instead of 10). The Prisma is a superb looking device (check  here for nice pictures), but I seeked for the Blitz considering the consistency of its concept. Strong Julio Kaplan programs are known for their advanced evaluation, full of promise for smart moves ; at the expense of the speed across the tree search. So, despite the program being able to focus on the most useful variations thanks to its evaluation skills, it needs time to decently analyse a position. Not the ideal way to play a speed chess game... To deserve the 'Blitz' name, the program has been reviewed, probably with Craig Barnes' support, with a much lightened evaluation for the benefit of its speed; and the brand new H8 has been chosen (it just reached the market). As a result, the program analyses 730 positions per second, as an order of magnitude. This is not outstanding, using the same hardware a GK2000 runs twice as fast, leveraging its Morsch program... But with regards to 95 positions per second the full-size version "B" program achieves in the Tandy 2150 (3Mhz 6502), thus an estimated 160 positions per second for the Saitek Turbo King with the very same "B" version (5Mhz 6502), the gain is glaringly obvious! (as for the "D+" version run by the Turbo King II, it achieves circa 310 nodes per second). By the way, the 1990 Saitek catalog states: 'The powerful new program on a brand-new advanced processor makes this our fastest-ever problem-solver'. More attractive to my opinion is the fact, this change results in a quite different play, deserving to add this program to my collection. In addition to the program and the processor, the chessboard is the third item made suitable for a blitz game: larger (25x25cm, to be compared to the Prisma or Turbo King 20x20cm) thus enabling easier quick moves, and above all it is an auto-response board: it automatically detects the pieces presence and moves, without requiring the user to press squares, and it features one LED per square to show its moves (as well displayed thanks to a 11 characters LCD window). The technology used is uncommon: leveraging the Hall effect instead of reed switches (those are toggled open or close according to the presence of a piece, equipped with a magnet). Here, each square hides two elements (copper loops), one acting as the magnetic field emitter, the other one as the receiver. Chess pieces are equipped with a non-magnetic metal ring, altering the field when put down on a square, and the receiver detects it. It works! This is not the end as far as innovation is concerned: this chess computer works without any keyboard, resulting in a clear design. The control interface features two wheels discreetly located on the right edge of the board. One wheel with notches to select the mode (level, options, game, info, ready, set-up), and one free wheel to set the value (or select a function) within the mode. The selected mode is simply printed on the first wheel (and readable thru the small rightmost window, labelled 'Menu') while the free wheel controls the LCD display on the left, labelled 'Data'. The free wheel has magnets inserted, the program detects them thanks to two reed switches, without any involved mechanical means. This unusual system is very efficient and comfortable with some of the modes, requiring repeated input, such as the setting or verification of a position. With other modes, it is not that convincing: a dedicated key enables direct access to a function; while a selection wheel only offers a sequential access... Besides, this system will no more be used in any other model. The example I bought suffered a failure there: it did not detect the direction the free wheel was rotated to. This did not prevent from playing, but it lessened the efficiency of the selection (for lack of roll-back), and the check of a position was not possible. Fortunately, both reed switches were accessible without too much dismantling; next to spotting the failing one (using an ohmmeter) I could replace it (using a MKA14103 reference, sold in twos for €2.78, so less than €5 including postal cost!) and my Blitz is now fully operational.


As does the Turbo King II displayed next page, the Blitz begins games in a bold way, leveraging strong playing skills once out of book, and up to the middle game included. On the whole, the family connections between both programs are obvious; to get convinced you just need to use the "Compare to" feature. So, the Turbo King II comparative endgame weakness (as much theoretical as practical) is as well present here; even more pronounced due to the less advanced evaluation function of the Blitz. Worth pointing out, the calculations power is slightly inferior despite the 30% faster processor; grouping this information with the better sighting of sacrifices opportunities, we can guess the Blitz program is a bit less selective than the Turbo King II one, resulting in a wider, but not as deep, vision.

Fidelity Excellence

Year: 1985
Programmer: Dan & Kathe Spracklen
CPU: 65C02 @3Mhz
ROM: 16Kb
Elo level: 1780 (1727 FIDE)
CMhz: 3
Rperf: 100%
KT: 1441
Square size: 1"

Bought brand new in 1985 with one of my first paychecks, retained up to now, it is my reference chess computer. A breakthrough shock when I realized its strength, with regards to my former Chess Challenger 7! Can't remind loosing to the CC7, can't remind winning to the Excellence... But no worry, I can find electronic opponents able to punish it... By constant playing I discovered a bug in its opening book, the Excellence loses a knight while playing these moves as white: 1.c4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.Nxe5? Okay, in this position I can win, but it is no fun. This computer has been sold in large number thanks to its price to playing level ratio; Fidelity has lowered the cost with the simplified design, featuring LEDs only for coordinates. I managed to perform its Khmelnitsky test leveraging its clone Fidelity Excel Display 3Mhz:


In my introduction page, I mentionned the feeling "a tough nut to crack" I had with this chess computer: it is quite revealed in the graph, with strong skills in opening and middlegame, a good positional game (strategy), so the program won't offer you easy ways a priori; and its ability to recognize threats, defend and counterattack whilst identifying tactical patterns will punish any ill-considered initiative. On another hand, its calculations power remains a weak point, partly due to the age of the processor, partly to the mainly "brute force" approach the program uses, not enabling deep scanning of variants. As for many chess computers from the same period, the materialistic way of playing is not in favor of sacrifices, and the program is not in a great mood for attacking (lacking an attack plan). It has a tendency to leverage its sound play while expecting your own mistakes. A reading I recommend: Scandien's page about chess engines playing style.

Systema Challenge

Year: 1994
Programmer: Gyula Horvath
CPU: H8 @20/2Mhz
ROM: 32Kb
Elo level: 1790 (1735 FIDE)
CMhz: 6.5
Rperf: 97%
KT: 1474
Square size: 1"

Yet another chess computer I chased for months, despite it exists in many variants, ten or so devices, launched from 1992 to 1996: its twin Krypton Challenge; and clones: CXG Sphinx (Accolade, Concerto, Legend, Legend II), Excalibur (Avenger, Legend II), Krypton (Comet, Regency). Clones feature the same hardware characteristics, the same software (or so), but different box shapings (and different brands!). I thus choosed the Systema Challenge... for the single reason I found it first, on UK eBay, bought complete with original packaging and mint condition, at €37.08 price! Anecdote: after months starvation, a few days after buying this one, another Systema Challenge showed up on UK eBay again... The reason I chased such a device lies in the authorship: Gyula Horvath. He is renowned as the author of Pandix, commercially  launched in 1993 using the product name ChessFriend, after achieving a few titles (amateur world champion in 1987, world PC software champion in 1988). However, very few contenders (7) attended these championships, and Pandix got these minor titles by achieving 5th rank. More significant to my opinion, and contemporary to the chess computers we are concerned with, is his great performance during 12th world micro-computer chess championship (Munich, 1993): 7th rank out of 28, achieving the same score points as did Mephisto Gideon, Quest (experimental Fritz), Kallisto and MChess ProX! With an impressively strong podium just ahead: Hiarcs, The King and Mephisto Genius 2! After ten years or so break, Gyula resumed working on Pandix in 2006, and was back to competition in 2009. Pandix achieved great results in 2013 in Yokohama, notably world chess software vice champion (every contender using the very same hardware) behind Hiarcs and on par with Junior. The climax of this saga is reached same year, Pandix being elected as the new Fritz 14 engine. Succeeding Frans Morsch is an impressive achievement!
The style of play developped by the 32K version powering the clones is renowned for being much positional, I even red someone calling it a procrastinator! Actually, initiative is not its strong point; and despite the H8 speed, its tactical ability is limited. Analysis modus enables displaying the on-going evaluation; noting a late score impact from a position is not uncommon, compared to its roughly same level computer opponent, displaying the impact a couple of half moves sooner. This is especially true during endgame, the Challenge does not anticipate enough pawns running for promotion. Some dedicated code is surely lacking in this area; in addition to a quite pronounced horizon effect at short time control. However, seven evaluation parameters are made available to the player to tune its playing style; and nine styles can be selected, from ultra-cautious to ultra-aggressive. Actually, these are nothing but presets of the seven available features; the default style and default features settings are expected to be the strongest ones. I nevertheless wanted to tune the excessively passive playing style, starting with trying to understand which parameter do have an impact; I used the Ketterling initiative test to achieve this. The test is based on playing the most passive way (usually moving a knight forth and back to initial position) while observing the number of moves the computer spends to mate (also an opportunity to observe the way it develops its pieces). I first set the features to extreme values (0 to 99), 10 being the standard value; to identify which feature did influence most, and using which direction for setting it. I used level 7 (2s) to restrain tactical influence:


Features 4 and 5 (aggressiveness from white and black, respectively) revealed the most effective with regards to expected initiative, and lowering them from 10 to 8 is enough to bring some drive. Then I checked this did not hurt the playing skills, using Spacious_Mind's 5 classic games (more details are provided with the Tiger Grenadier, hereafter). I concluded  facing the Challenge using this setting to four opponents, 1773 average Elo level, eight games; but it poorly achieved 0,5 points out of 8! I noticed more particularly an excessive inclination to move its rooks on open files (or semi-open ones); to pin opponent's pieces onto the queen without lasting advantage, to let its pawns get doubled (adding difficulty to its already weak endgame) and to sometimes neglegt to play useful threats on enemy king, lacking tactical forecasting computation. So, I tuned the corresponding features, reaching hereafter values after proceeding by trial and error:
  F1 (pin against enemy queen) = 9, lowered by 10%
  F2 (pawn structure around king) = 10, unchanged
  F3 (king attack by pieces) = 12, raised by 20%
  F4 & F5 (white & black aggressiveness) = 8, as stated above
  F6 (doubled pawns) = 11, raised by 10%
  F7 (rooks on open/semi-open files) = 9, lowered by 10%
I checked this combination using Spacious_Mind's test, it achieved close to 60 Elo points more than default setting, with an interesting balance: four out of five games scored higher than the standard, and the single lower score is less than 30 points below standard result.


I concluded with playing again the 8 games facing the same opponents, this occurrence the Systema Challenge achieved 4 points out of 8; thus a performance equal to the average Elo of the opponents (1773). And the playing style was more balanced and straightforward, to my opinion. Setting approved!
The boxing itself does not invite much comments, sensor squares and keyboard are sensitive enough, and small LEDs on coordinates do help as the LCD display is a bit small and out of sight, too deeply fitted. It nevertheless enables browsing the menu, and displaying rotating information while the program is thinking (main variant, score, thinking time). I point out an unusual feature: 'Why Not?' enabling the player to display the program refutation to whatever move it did not play.


Khmelnitsky test: this knowledge profile is rather close to the Fidelity Excellence's one, slightly weaker with regards to the standard endgame positions and strategy, on another hand slightly stronger at tactics and counterattack. Like the Excellence, it is a "brute force"-type program ("A" strategy according to Shannon): it does not selectively extend its search, thus limiting its ability to deeply calculate. Each position from the test has been analyzed twice, with and without the above setting; the special setting slightly decreases the global score (-33 points, leading to the very same score as the Excellence, being 1441). This was not unexpected, the default setting is supposed to be the strongest; 7 domains loose some points, 4 are unaffected, and only one performs better: strategy (well really, +127 points!). After having analyzed 100 diagrams from the test, I can tell the changes in the scores due to the setting remained low, most usually within the order of magnitude from 5 to 15 centipawns; and only triggered any difference in positions with still a load of pieces (from the end of the openings to the early middle game).



next category: 'strong club player, class A level' chess computers
previous category: 'weak club player, class C level' chess computers
back to 'my chess computers'