Author: Keith Evans
Date: 16:36:13 03/20/02
Go up one level in this thread
On March 20, 2002 at 12:23:53, Steve Maughan wrote:
>Interesting!! I had no idea that ChessBase were behind the project. If it's
>true that adding extra knowledge has no cost (which I find hard to believe!!)
>then it's a little strange that they chose Donninger as the programmer since
>Nimzo is IMO not the most knowledge rich program - I would have thought
>Meyer-Kahlen would have been better. Nevertheless, I'll watch with interest!
>
>Steve
It's not completely true that adding knowledge to a chess chip doesn't
have a cost. There are basically two problems that you will run into:
1 - you exhaust the resources available in a chip and there's no room
to add any additional logic
2 - by adding logic you may reduce the maximum operating frequency of
a chip (maybe fanout on certain critical nets increased, or
you ran into problems with routing. In a Xilinx FPGA routing
delays are often a problem.) It's sometimes possible to add
pipelining to a chip to increase the operating frequency, but
then you'll have larger latencies which you may not be able to hide.
When using a Xilinx FPGA both of the above are important.
Depending on how full your FPGA is, it could be correct to say that a
certain amount of knowledge could be added with zero cost. And if you
run into problems when doing this, but you're willing to pay for a larger
or faster Xilinx part, then you might be able to add knowledge without
compromising the maximum operating frequency. This will not always be the
case.
I think that when people mention "cost" in this context they are always
referring to the maximum operating frequency.
Keith
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.