I think its been explained here how Crafty has a somewhat lower NPS
speed than other programs out there. I've recently used Arena in
tournament mode to face off Ruffian 1.0.5 and Crafty 19.13
(timecontrol of 40 moves in 35 min) and not only were the results
steeply in favor of Ruffian but it was clear that Ruffian was
searching up to 70% more nodes than Crafty. I can't help but think
that Ruffian's speed is a factor in its success. What is it about the
algorithms in Crafty that hold it back in the NPS department?

I compiled Crafty will all the optimizations I could, while Ruffian
came as a binary of course.

jake johnson says...

I think its been explained here how Crafty has a somewhat lower NPS
speed than other programs out there. I've recently used Arena in
tournament mode to face off Ruffian 1.0.5 and Crafty 19.13
(timecontrol of 40 moves in 35 min) and not only were the results
steeply in favor of Ruffian but it was clear that Ruffian was
searching up to 70% more nodes than Crafty. I can't help but think
that Ruffian's speed is a factor in its success. What is it about the
algorithms in Crafty that hold it back in the NPS department?

I compiled Crafty will all the optimizations I could, while Ruffian
came as a binary of course.

I wouldn't be surprised at all if I could speed up Crafty by 10X.
It would involve running it on a special PC I use that has dual
port RAM and a second processor that profiles where the program
spends it's time, followed by months of painstaking assembly language
programming to optimize the sections that eat the most time. I have
managed projects where this was done and got 5X to 15X speedups.

There is also the possibility that Ruffian has a secret algorithm
that makes it faster. Remember, Ruffian can stael all of Crafty's
algorithms, but the reverse is not true.






--
Guy Macon, Electronics Engineer & Project Manager for hire.
Remember Doc Brown from the _Back to the Future_ movies? Do you
have an "impossible" engineering project that only someone like
Doc Brown can solve? My resume is at http://www.guymacon.com/

Guy Macon http://www.guymacon.com wrote in
:

jake johnson says...

I think its been explained here how Crafty has a somewhat lower NPS
speed than other programs out there. I've recently used Arena in
tournament mode to face off Ruffian 1.0.5 and Crafty 19.13
(timecontrol of 40 moves in 35 min) and not only were the results
steeply in favor of Ruffian but it was clear that Ruffian was
searching up to 70% more nodes than Crafty. I can't help but think
that Ruffian's speed is a factor in its success. What is it about the
algorithms in Crafty that hold it back in the NPS department?

I compiled Crafty will all the optimizations I could, while Ruffian
came as a binary of course.

I wouldn't be surprised at all if I could speed up Crafty by 10X.
It would involve running it on a special PC I use that has dual

For a second, I thought you were going to say, "It would involve stubbing
out the evaluation code."

(jake johnson) writes:

I think its been explained here how Crafty has a somewhat lower NPS
speed than other programs out there.

Not really. It is probably slower than Fritz, Junior and Ruffian, but
faster than most other engines, including top programs like Hiarcs,
The King and Shredder.

You simply can't make any conclusions about an engine's strength by
looking at the NPS count.

--
Tord Romstad

In article , Guy Macon
http://www.guymacon.com wrote:


There is also the possibility that Ruffian has a secret algorithm
that makes it faster. Remember, Ruffian can stael all of Crafty's
algorithms, but the reverse is not true.

It is possible to speed up a chess program by simply narrowing its search,
thus depening it for a set amoount of time. Many programs have this
option, including Fritz and the Palm version of Hiarcs.

Of course the price paid is that the program will not analyze some good
move deep enough to find out that it is a good move, so there is an
optimum in there that depends on the position.

So depth of search alone without knowing how much width was sacrificed can
be misleading, as can the number of nodes, although all those are not
worthless indications.

Henri