How many moves did you simulate with each?  Did you have the same
position/rack in both?  The move generation time is affected by both
board and rack and so simulation times will vary with them.

I simulated the top 10 choices for 300 iterations at 2 ply in 3:15 on
my laptop at some random position.  

IMHO, there is 2-5X of performance improvement in quackle to be had
without dropping any moves (based on general profiling, not anything
I've actually implemented).  I'm not sure when/if we will achieve that
though.

Matt

--- In [email protected], "David Webb" <[EMAIL PROTECTED]> wrote:
>
> I have downloaded Quackle 0.94 but SOWPODS simulation still seems
very slow.  I simmed a random rack for 300 iterations, comparing Maven
(using default looking ahead one move per side) and Quackle (using
default two plies), and got the following results:
> 
> Maven (using Windows ME) - 50 seconds
> 
> Maven (using Windows XP) - 2 minutes 34 seconds
> 
> Quackle (using Windows XP) - 18 minutes 6 seconds
> 
> Is Quackle's comparative slowness due to 1) my computer 2) not
comparing like with like 3) Quackle just being slower or 4) something else
> 
> In order to be useful for analysing whole games in a reasonable
time, a 300 iteration sim needs to take at most 90 seconds, in my
opinion.  Is this likely to be achievable in a future release of
Quackle?  (If the comparative slowness is due to my computer these
speeds may of course be achievable already).
> 
> I appreciate that speed can be increased by deleting obvious
non-optimal moves but I like to see just how bad non-optimal moves
are.  Also I may delete an obviously non-optimal move which is in fact
optimal for reasons I haven't appreciated and that would be a valuable
learning opportunity missed.
> 
> One final point is that some years ago people commented that Maven
was slower than they wanted.  One response was that Maven was built
for old computers and did not take advantage of features in modern
computers that could enhance speed.  This led to an expectation
amongst some people, including myself, that a replacement for Maven
would offer faster simming. Were we mistaken in this belief?
> 
> I hope this email does not come across as critical.  I am delighted
that a replacement for Maven is being developed, particularly one that
proxy sims for winningness.  Good luck in Toronto.
> 
> David
>



Reply via email to