What was the top Chess engine before it was Stockfish, Komodo or Houdini?

[Rebel]

Today we have
Phoenix NL
DGT NL
Pewatronik Switzerland
Millennium D


So the dedicated chess computer companies survived.

Of course novag, fidelity, saitek, Mephisto, cxg, ...
are all dead .

I would say the problems of dedicated chess computers began earlier. With the 486.

Yes.

The 486 was a breakthrough.

[S.Taylor]

2006 Rebel - win over world champion (Kramnik) in a 6 game match using Intel Pentium PC

I may have wished so, but it was Fritz.

At that time, genius was stronger than fritz but got less attention.

[yurikvelo]

There is confusion in this topic on term "dedicated chess hardware"

Electronic chess boards, like DGT NL, Pewatronik etc - use general purpose low-end x86 or ARM processors. They were not build to beat Kasparov or to beat Stockfish @ TCEC-grade hardware.
They have dedicated user interface, not dedicated CPU and not even high-end general purpose CPU.

I mentioned dedicated hardware for top-performing engines. Dedicated electronic chips, specially designed to run chess calculation faster than general purpose off-shelf CPU from Intel, AMD, Sun etc


Hydra used dedicated XilinX based VirtexV1000E FPGA (Field-Programmable Gate Array), which performed 3-ply depth eval/search faster than any CPU.
Internal 140 test games of Hydra with only 4 processors against Fritz8 and Shredder8, playing on a Pentium 2.4 GHz PC, showed an advantage of more than 110 ELO point

IBM Deep Blue also employed dedicated FPGA chips to reach speed of 200 MN/sec in 1997 and 100 Mn/sec in 1996.

[Vinvin]

2006 Rebel - win over world champion (Kramnik) in a 6 game match using Intel Pentium PC

I may have wished so, but it was Fritz.

I didn't see that glitch
All famous Man Computer Matches here : https://www.chessprogramming.org/Tourna ... er_Matches
And all famous Man Computer Tournaments here : https://www.chessprogramming.org/Tourna ... ournaments

[Vinvin]

IBM Deep Blue also employed dedicated FPGA chips to reach speed of 200 MN/sec in 1997 and 100 Mn/sec in 1996.

Here, I'd like to add that Deep Blue consist of 30 nodes, each node composed with 16 Chess Chips (FPGA chips), so 480 Chess Chips overall.
This system scored a perf around 2800 Elo in 1996-1997.
For reference, the current strength of Stockfish 9 on 1 core Athlon 64 X2 2.4 GHz is around 3370 (570 Elo over Deep Blue).

[Colin-G]

I notice that Chess Genius by Richard Lang has been mentioned in this thread.
Does anyone know what the elo is for the latest PC version 7.2?
Is there a reason it is not included in the CCRL list?

[tpoppins]

Here, I'd like to add that Deep Blue consist of 30 nodes, each node composed with 16 Chess Chips (FPGA chips), so 480 Chess Chips overall.
This system scored a perf around 2800 Elo in 1996-1997.
For reference, the current strength of Stockfish 9 on 1 core Athlon 64 X2 2.4 GHz is around 3370 (570 Elo over Deep Blue).

Vince, this comparison makes little sense because it is based on two fallacies.

The first one is comparing Deep Blue's performance rating to SF9's Elo rating; the other is comparing a(n imaginary) FIDE rating (Deep Blue never had one) to CCRL ratings. Elo ratings are only valid for players in the same playing pool.

DB only ever played Kasparov - six games, error bars measuring in hundreds; Kasparov played (also only six games with resultant huge error bars) a Deep Junior version that is two years older than the oldest Junior version on our lists, and of course that version would be too low-rated to ever get a crack at playing SF9 even if we had it. You could try and extrapolate DJ 2003's rating from the subsequent versions and other sources but the connection between DB and SF9, if one can be said to exist at all, is nebulous at best. With all the extrapolation, guesses and error margins taken into account you'd be as justified in quoting a 1000 Elo or a 150 Elo difference as quoting your 570 Elo.

I notice that Chess Genius by Richard Lang has been mentioned in this thread.
Does anyone know what the elo is for the latest PC version 7.2?
Is there a reason it is not included in the CCRL list?

Strength of Chess Genius Classic
I suppose the reason it's not on CCRL is that it runs in its own proprietary GUI only, being neither UCI- nor CECP-compliant. Still, with CEGT having tested it, perhaps it's only a matter of time of it getting listed on CCRL as well.

[yurikvelo]

Deep Blue rating has very wide bars due to many facts:

* very low game count = 6
* Kasparov used Pentium 166MMX with some strong engine of that time (don't remember which one)
* specially for this match, Kasparov didn't play naturally at his best. He had prejudice on computer play style, based on his previous experience with other computers and decided to play what he believed "anti-computer" moves.
Kasparov attributed 44... Rd1 to "superior intelligence", though it is 17th worst out of 25 legal moves and forced M17 for machine

* during most of human-maching matches - special adjudication rule was used: "resign as early as possible / offer draw as early as possible". Humans are not prone to blunders and to fatigue, so this rule favours humans a lot.

here are few multiPVs from Deep Blue matches:

https://pastebin.com/mD0jGM2L (1997, Game 2. Kasparov resigned in draw position)
https://pastebin.com/LS8hEbjx
https://pastebin.com/SpmX7123
https://pastebin.com/saXJcQVP
https://pastebin.com/2zpUJfDJ
https://pastebin.com/K2qzrRt6

[Vinvin]

Vince, this comparison makes little sense because it is based on two fallacies.

The first one is comparing Deep Blue's performance rating to SF9's Elo rating; the other is comparing a(n imaginary) FIDE rating (Deep Blue never had one) to CCRL ratings. Elo ratings are only valid for players in the same playing pool.

History of humans vs computers games shows that the rating lists are close to human strength. See dozens of results here :
https://www.chessprogramming.org/Tourna ... ournaments
https://www.chessprogramming.org/Tourna ... er_Matches

So, 2800 Elo for Shredder 12 on 1 CPU around 2 GHz seems realistic. Some lists are a bit higher.
http://www.inwoba.de/index.html
http://www.computerchess.org.uk/ccrl/40 ... t_all.html
http://www.cegt.net/40_40%20Rating%20Li ... liste.html
https://ssdf.bosjo.net/long.txt

DB only ever played Kasparov - six games, error bars measuring in hundreds; Kasparov played (also only six games with resultant huge error bars) a Deep Junior version that is two years older than the oldest Junior version on our lists, and of course that version would be too low-rated to ever get a crack at playing SF9 even if we had it. You could try and extrapolate DJ 2003's rating from the subsequent versions and other sources but the connection between DB and SF9, if one can be said to exist at all, is nebulous at best. With all the extrapolation, guesses and error margins taken into account you'd be as justified in quoting a 1000 Elo or a 150 Elo difference as quoting your 570 Elo.

Considering the Kasparov-DB match, I always take the 12 games into account. But DB also played games versus GMs in 1993, scored a slight + perf, that indicates it was around 2600-2700 at this time.
http://www.chessgames.com/perl/chessplayer?pid=29912

"570 Elo" is an approximation as all ratings are rough approximations.