Re: computer-go: FPGA

[engels@xxxxxxxxxx (Andre Engels)]

jliu wrote:

> I think when computer speed-up, it speed-up linearly, while the difficulty
> of search explodes exponentially with respect to depth.

No, computer speed-up is also exponentially. It's just that the difficulty
of search explodes exponentially with such a large base number. If we were
to do a 'stupid' full width search, this grows exponentially in the number
of ply with base number 300 or so (for the early middle game). Computers
get twice as fast in (if I remember correctly, which I probably don't)
1 1/2 years, which means that we get one ply deeper in about 12 years.

Of course with better methods the degree of branching in the actual search
tree can be brought down drastically, but still, if you have a branching
factor 20, an extra ply will take more than 6 years, whereas it is now
assumed that for supercomputers the 'ceiling' of what is technologically
possible will be reached in 8 years or so.

Thus, don't expect a gain of more than 1 ply through computer speed-up
in the foreseeable future.


-- 
Andre Engels, engels@xxxxxxxxxxxxxxxxx
telephone: +31-40-2474628 (work), +31-6-27174384 (mobile)
http://www.win.tue.nl/~engels/index_en.html

A child is not a glass that is filled, but a fire that is set ablaze.
						- Maria Montessori