2005年世界{gj}“九头蛇怪” Hydra运行于阿联酋迪拜的一个64处理器阵列上的32个处理器上。它通过因特网将招法传至毕尔巴鄂。当设计人员在被问及为什么不使用64个处理器时说，Hydra的程序设计是针对32处理器优化的。每个处理器都具备专门的硬件增强，其中Hydra的运算核心部分(搜索和端节点评估)运行于一个超快速的现场可编程门阵列(FPGA）芯片上。设计人员介绍说目前他们正在打算更新程序，按照128处理器进行优化，届时这个“九头蛇怪”将更强大更可怕。


The Hydra chess computer looks like it is going to be the
new world computer chess champion, and it is getting close
to being the best chess player, human or computer.

What does this have to do with embedded systems? Hydra is
FPGA-based, not PC-based - carrying on the tradition of
Deep Blue and swinging the pendulum back to hardware.

From what I have been able to gather, The Hydra chess
computer runs on 8 Alpha-Data PCI ADM-XRC Xilinx Virtex-EM
V405E-V812E FPGA Cards.

It was programmed by Chrilly Donninger in two languages;
C for the high-level behavioral prototyping and specification
language and Verilog for the low-level RTL implementation
language. Chrilly has also been working on a computer that
plays a good game of Go using pattern recognition and possibly
cellular automata.

(The Hydra project was originally known as the Brutus project.)
Hydra currently runs on a 32 node Xeon cluster, with a total of 64 gigabytes of RAM. It evaluates about 150,000,000 chess positions per second, roughly the same as the much older Deep Blue, 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 (9 moves by each player), deeper than Deep Blue, which only evaluated to about 12 ply on average. Hydra's search uses 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[citation needed].

The Hydra computer is physically located in Abu Dhabi, in the United Arab Emirates, and is usually operated over a high speed optical fiber based network link.

{zx1}版的Hydra叫作HYDRA 64 CL.

HYDRA is the chess monster of UAE possessing ultimate dominance in the computer chess world.

HYDRA, under the patronage of PAL GROUP, originally appeared in four versions:  Orthus, Chimera, Scylla and then the final HYDRA version-the strongest of them all.  Multiple computers each with its own FPGA acting as chess coprocessors, enables HDYRA to search enormous numbers of positions per second, making each processor more than ten times than an unaided computer.

It runs on a 64CL and evaluates about 200,000,000 chess positions per second, roughly the same as the much older Deep Blue, but with several times more overall computing power.  Whilst FPGAs generally have a lower performance level than ASIC chips, Moore’s Law allows modern-day FPGAs to 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 (9 moves by each player), deeper than Deep Blue, which only evaluated to about 12 ply on average.  HYDRA’s search uses alpha-beta pruning as well as null-move heuristics.  The extra search depth over Deep Blue is due to its use of more modern type B forward pruning techniques that are slightly less perfect but generally play better due to the greater search depth these techniques permit.

Under the stewardship of Dr. Chrilly Donninger, HYDRA R&D is currently developing a 64CL new version.
除了以FPGA加速硬件设计为特色Hydfra，
还有一个叫Rybka的软件，运行于cluster之上。

For those of you interested in computer chess you need to check this stuff out...

Hardware expert Lukas Cimiotti (nick "Kullberg" in the forum) built the ultimate chess computer system, a cluster consisting of four Nehalem dual socket Intel Xeon W5580 (=8 cores each) computers 2.93 - 3.33 GHz and 5 Nehalem quads 4.2 GHz.
The cluster uses a private version of Rybka (Rybka 3 is currently the strongest chess software in the market), which, besides of supporting clusters, is stronger than Rybka 3.

So far the cluster lost only one game (of 24) against a human + computer team!