Sloppy 0.2.0 released

[Jouni]

About bitbases: also Toga and bright are worse in EET suite with full 5 piece set and better with 3-4 set. Any good explanations??

Jouni

[David Dahlem]

Hi David,

I have Sloppy 0.2.0 installed in Arena, but i can't seem to adjust the hash size. In the config file, i have this entry ...

# Hash table size in megabytes
# hash = 256

In your 'config' example the hash symbol '#' should be removed to activate the option e.g

Code: Select all

# hash = 256     = ignored
  hash = 265     = used

rgds,
Jim.

Thanks Jim. That solved the problem. I suppose not enough coffee early this morning was responsible for my dumb mistake.

Regards
Dave

[Werner]

Hi Jim,
on my Athlon 64x2 there is no speed difference to be seen between auto paralized version or normal version. Does this only work on Intels?
The bench is even running a bit faster on the nomal version and inside taskmanager I always see 50 %

[Jim Ablett]

Hi Werner,

Hi Jim,
on my Athlon 64x2 there is no speed difference to be seen between auto paralized version or normal version. Does this only work on Intels?
The bench is even running a bit faster on the nomal version and inside taskmanager I always see 50 %

Hi Werner,

Gains will only be seen when the compiler finds suitable loops for parallelizing, and
whether the loops it converts are the ones that affect search speed,
so it's very much dependent on the source code, and results will differ from
engine to engine. The parallelized builds should be seen just as test versions for you to try out.

regards,
Jim.

[ilari]

About bitbases: also Toga and bright are worse in EET suite with full 5 piece set and better with 3-4 set. Any good explanations??

Jouni

I think the worse rating in general would be because 5-men egbbs take so much time to probe, or perhaps there are bugs in the egbb library. But if the EET suite requires the engines to find mates, then it's only natural that egbbs make things worse due to the fact that they don't tell the distance to mate. There are many positions where Sloppy wins faster or loses slower without bitbases, but it doesn't really affect the end result and has no impact on the rating.

[pedrox]

Sloppy consultation bitbases in qsearch and in the search, it is difficult to adjust this in the search.

In the default configuration, Sloppy seem to be charging 4 men in memory and 5 men in hard disk ?. When, it would be advisable to use a more cache_size, instead of using 4 MB, 32 MB (the minimum as Daniel said is 16 MB)

[ilari]

In the default configuration, Sloppy seem to be charging 4 men in memory and 5 men in hard disk ?. When, it would be advisable to use a more cache_size, instead of using 4 MB, 32 MB (the minimum as Daniel said is 16 MB)

Actually the 5-men egbbs are completely disabled by default. Other than that, the defaults are pretty much the same that Scorpio has. With 3-men and 4-men bitbases in memory, I found that 4MB of cache is enough. But of course everyone can use their own settings and see if they can find something that works better.

[gerold]

Downloads and more info here: http://koti.mbnet.fi/~ilaripih/sloppy/

As it says on the site, I'm making (and already made) some pretty big changes to Sloppy, so I wanted to get one more stable and fully functional version out before things get messy. Sloppy 0.2.0 is that release. I call it a small update, but it's still significantly stronger than the previous version. This version also shouldn't have such a huge problem with the fast time controls that Tony Thomas uses.

The biggest new feature is support for Scorpio's endgame bitbases. It was pretty easy to implement but required a lot of testing. Based on the tests I have concluded that 4-men bitbases (loaded in RAM) help Sloppy with a lot of extra ELO points. 5-men egbbs on the other hand aren't doing much good, and if they're loaded in RAM Sloppy actually seems to get weaker.

As always, feedback, bug reports, and faster builds are welcome.

Quick check. 2/12
Ruffian 101 vs.Sloppy-0.2.0
Ruffian 23/45
Sloppy 22/45
Looks like they are close in elo.