Hydra (chess)

Hydra was a chess machine, designed by a team with Dr. Christian "Chrilly" Donninger, Dr. Ulf Lorenz, GM Christopher Lutz and Muhammad Nasir Ali. Since 2006 the development team consisted only of Donninger and Lutz. Hydra was under the patronage of the PAL Group and Sheikh Tahnoon Bin Zayed Al Nahyan of Abu Dhabi. The goal of the Hydra Project was to dominate the computer chess world, and finally have an accepted victory over humans.

Hydra represented a potentially significant leap in the strength of computer chess. Design team member Lorenz estimates its FIDE equivalent playing strength to be over Elo 3000, and this is in line with its results against Michael Adams and Shredder 8, the former micro-computer chess champion.

Hydra began competing in 2002 and played its last game in June 2006. In June 2009, Christopher Lutz stated that "unfortunately the Hydra project is discontinued." The sponsors decided to end the project.

The Hydra team originally planned to have Hydra appear in four versions: Orthus, Chimera, Scylla and then the final Hydra version – the strongest of them all. The original version of Hydra evolved from an earlier design called Brutus and works in a similar fashion to Deep Blue, utilising large numbers of purpose-designed chips (in this case implemented as a field-programmable gate array or FPGA). In Hydra, there are multiple computers, each with its own FPGA acting as a chess coprocessor. These co-processors enabled Hydra to search enormous numbers of positions per second, making each processor more than ten times faster than an unaided computer.

Hydra ran on a 32-node Intel Xeon with a Xilinx FPGA accelerator card cluster, with a total of 64 gigabytes of RAM. It evaluates about 150,000,000 chess positions per second, roughly the same as the 1997 Deep Blue which defeated Garry Kasparov, but with several times more overall computing power. Whilst FPGAs generally have a lower performance level than ASIC chips, modern-day FPGAs run about as fast as the older ASICs used for Deep Blue. The engine is on average able to evaluate up to a depth of about 18 ply (nine moves by each player), whereas Deep Blue only evaluated to about 12 ply on average. Hydra's search used alpha-beta pruning as well as null-move heuristics [1]. The extra search depth over Deep Blue is due to its use of more modern type B forward pruning techniques that can miss some possibilities but generally play better due to the greater search depth permitted.

...
Wikipedia