Re: [computer-go] Computer Go hardware

>Your system is nothing short of a supercomputer on its own.

What means "nothing short of" (have not found this phrase in my English-dictionary)? If it means, it is no supercomputer, I disaggree. I am working currently on a 32/64-processor version. The machine will calculate 300 Million positions/sec peak. 1 position is - on the C-simulator - 40.000 instructions. Makes (if I have calculated right) 120 Tera-MIPS.

32/64 is probably the greatest number of processors which can be reasonably used. The parallel speedup drops rapidly with more processors. The Japanese earthquake-machine would be certainly slower/less efficient than the next Hydra FPGA-Cluster. In Go the number of usefull processors is probably somewhat higher. But even here: On a >= 1024 processor machine each processor is mainly busy with communication. I got the code from the massive parallel Zugzwang chess programm. I had problems to find the chessprogramm at all. It was a big communication loop and only in the communication-idle time was a small step of chess-computation done. This is the only way to keep such a system running. The nodes/per CPU were in Zugzwang extremly low.

Note: I call it a 32/64 processor machine, because it has 32-physical nodes, but each FPGA contains 2 chess cores.

>Many of us are glad that computer Go has not matured to the stage that it's dominated by organized academics. It will happen soon or later.

I am not at all an organized academic. I am a freelance professional chess programmer who has developed with very little money from ChessBase the Brutus-FPGA-Machine. Since beginning this year the project is called Hydra. Hydra is sponsored by Pal-Computing from the United Arab Emirates.

Computer-chess has indeed become much too professional/serious. But this rivalry has also triggered a remarkable development in playing strenght. I do not like this atmosphere and I have found the Computer-Go-olympiad in Graz 2003 much more enjoyable than the chess-WC.

Chrilly