Since I could not find Houdini 1 in any rating list, I conducted a small test match over 20 games between the first Houdini version from May 2010 and the chess program Critter 0.90 by the Slovakian programmer Richard Vida which was published half a year later. I used the CCRL standard time control of 4 minutes for 40 moves (64-bit, 1CPU) which was usual until 2019. The final result was 5-3 with 12 draws in favour of Houdini 1. Since Critter 0.90 64-bit has an ELO rating of 3049 in the CCRL 40/4 list, the corresponding ELO rating of Houdini 1 is somewhere around 3083. This is in the range of Rybka 3's rating of 3078 ELO. Of course, significantly more test games should have been played, but a few ELO up or down are not important here. I just wanted to check if the first version of Houdini already had a remarkable strength. The small test match against Critter confirmed this assumption.
Houdini 3 64-bit 4CPU
DOWNLOAD: https://tinurli.com/2vFkEb
Robert Houdart was born in 1968. He is a computer scientist and professional software developer with a FIDE ELO rating of 2101. (e) These are good qualifications to optimize a chess program more and more. The constant ongoing development of Houdini 1 soon paid off: Houdini 1.03a, released two months later in July 2010, had an ELO rating of 3116 ELO in the CCRL 40/4 list (64-bit, 1CPU). The last free Houdini version 1.5a from January 2011 gained 54 more ELO points and thus reached 3170. For comparison: The strongest Rybka version 4.1 had an ELO rating of 3108, while the commercial Houdini versions 2, 3 and 4 climbed up to 3202, 3217 and 3240 ELO. In the three and a half years between May 2010 (Houdini 1) and November 2013 (Houdini 4) Robert Houdart was thus able to improve Houdini from about 3083 ELO to 3240 ELO (64-bit, 1CPU). This is a plus of almost 160 ELO. Running on four processors (4CPU) Houdini 4 even reached 3332 ELO. (f)
(f) In the 40/40 rankings, the earlier Houdini version 1.03a was not tested, but in 40/4 it was, so the ELO ratings in this article refer to a time control of 4 minutes for 40 moves (repeating) and to the 64-bit versions using 1 CPU. With 4 processors (4CPU) the ELO ratings are correspondingly higher. With Houdini 6, for example, this would mean 3517 ELO (4CPU) compared to 3444 or 3445 (1CPU). Unfortunately, the proven 40/4 time control is no longer used by the CCRL testers for several months. Instead they now play blitz games with 2 minutes for the whole game + 1 second increment on faster hardware. The last archived 40/4 rankings I could find are from October and November 2019 respectively:
Da Houdini 1 in keiner ELO-Rangliste zu finden war, veranstaltete ich selbst einen kleinen Testwettkampf über 20 Partien zwischen dieser ersten Houdini-Version und dem ein halbes Jahr später erschienenen Schachprogramm Critter 0.90 des slowakischen Programmierers Richard Vida. Zugrundegelegt wurde die bis einschließlich 2019 übliche CCRL-Bedenkzeit von 4 Minuten für je 40 Züge (64-bit, 1CPU). Das Endergebnis lautete 5:3 bei 12 Remisen zugunsten von Houdini 1. Da Critter 0.90 auf 64-bit mit einem Prozessor eine ELO Zahl von 3049 hat, ergab die ELO-Auswertung für Houdini 1 eine ELO-Zahl von 3083, gemessen am CCRL-Standard 40/4. Das liegt im Bereich von Rybka 3, der auf 3078 ELO kommt. Natürlich müssten deutlich mehr Testpartien gespielt werden, aber auf ein paar ELO rauf oder runter kommt es hier nicht an. Ich wollte nur überprüfen, ob die erste Version von Houdini tatsächlich schon außerordentlich stark war. Der kleine Testwettkampf gegen Critter bestätigte jedenfalls diese Annahme.
Robert Houdart ist Jahrgang 1968 und hat eine FIDE-ELO-Zahl von 2101. (e) Von Beruf ist er Computerwissenschaftler und professioneller Software-Entwickler. (f) Beste Voraussetzungen also, ein Schachprogramm immer weiter zu optimieren. Die beharrliche Weiterentwicklung von Houdini 1 trug schon bald reiche Früchte: Die zwei Monate später, im Juli 2010 erschienene Version Houdini 1.03a brachte es in der CCRL 40/4-Liste bereits auf 3116 ELO-Punkte (64-bit, 1CPU), während der letzte kostenlos erhältliche Houdini 1.5a vom Januar 2011 mit 3170 noch einmal satte 54 ELO-Punkte zulegte. Zum Vergleich: Die stärkste Rybka-Version 4.1 hatte eine ELO-Zahl von 3108. Die von da an kommerziellen Houdini-Version 2, 3 und 4 steigerten sich weiter auf 3202, 3217 bzw. 3240 ELO-Punkte. In den dreieinhalb Jahren zwischen Mai 2010 (Houdini 1) und November 2013 (Houdini 4) konnte Robert Houdart damit Houdini von schätzungsweise 3083 ELO auf 3240 ELO verbessern (64-bit, 1CPU). Das ist ein Plus von fast 160 ELO-Punkten. Auf vier Prozessoren (4CPU) erreichte Houdini 4 sogar 3332 ELO. (g)
(g) In den 40/40-Ranglisten wurde die frühere Houdini-Version 1.03a nicht getestet, in 40/4 aber schon, so dass sich die ELO-Zahlen in diesem Artikel auf die mit einer Bedenkeit von 4 Minuten für je 40 Züge ermittelten ELO-Zahlen beziehen (64-bit, 1CPU). Mit 4 Prozessoren (4CPU) sind die ELO-Zahlen entsprechend höher. Bei Houdini 6 z. B. wären das 3517 ELO (4CPU) gegenüber 3444 bzw. 3445 (1CPU).
Provided binaries are for 64-bit Windows. Use the NOPOPCNT version for Intel processors before 2008 or non-Intel processors. Binaries have only been tested on 64-bit Windows and Intel processors.See README for more information.
Although Deadline marks the task as complete. It puts out an error that is not caught:STDOUT: [Redshift] Output file locked or path not available for writing. Skipping frame. (X:/17110093_achthamburg_researchanddevelopment/06_vfx/01_seq/deadline_test_190208/houdini/render/win2linux_env_job/deadline_test_190208_v001.Redshift_win2lin.0002.exr.lock)
Amazon EC2 T4g instances are EBS-optimized for enhanced networking, and each contain a custom AWS Graviton2 processor with a 64-bit Arm core. T4g instances deliver up to 40% better performance for the price than T3 instances and are appropriate for a broad set of burstable general purpose workloads, including microservices, low-latency interactive applications, small and medium databases, virtual desktops, development environments, code repos, and business-critical apps.
Amazon EC2 M6g instances are driven by 64-bit Neoverse Arm-based AWS Graviton2 processors that deliver up to 40% improvement in price and performance beyond current generation M5 instances and enable a balance of compute, memory, and networking resources to support a broad set of workloads.
CCRL Rating: 3232CEGT Rating: 3098Critter is the UCI chess engine available for Windows, Mac, Android, and Linux. You can use it for private purposes only. It was initially written in Delphi but later converted to C++ using Bitboard technology. This was done to enhance its performance on 64-bit processors.
CharlesRobert you certainly did a fine job with Houdini.It is very strong and is beating Rybka 4 on my machine.What I don't understand is why not make the simple statement: I took the ippolit sources, studied them, and made changes to produce what we have now. Robert HoudartBecause the simple fact is that this is not an accurate description of how Houdini was created....As a matter of fact, Houdini would have been written in Delphi (Object Pascal) if it weren't for the lack of a 64-bit Delphi compiler until next year. I'm a far greater Delphi expert than C++ expert. 2ff7e9595c
Comments