TalkChess.com

michiguel wrote:Then again, at fast pace when HT replacement is not critical.

An interesting point. Do you think that when testing at fast time controls that caches should be reduced so as to work approximately the same as at normal time control? Do you do this in your testing?

Dirt wrote:

michiguel wrote:Then again, at fast pace when HT replacement is not critical.

An interesting point. Do you think that when testing at fast time controls that caches should be reduced so as to work approximately the same as at normal time control? Do you do this in your testing?

I think I may not be a huge deal, but it is a detail that is worth paying attention to. I use 8 MB of hash tables in my own testing if I remember correctly (I have the data at home).

Miguel

Dann Corbit wrote:The new stockfish is really excellent against closed and semi-closed positions. It's very nice to have another option to analyze these positions.

I have tried both 1.7 and 1.7.1 in few closed midgame positions.
Almost all the time, evaluation value was bigger than what Rybka
or Naum or Shredder or other engine reports. And it changes very
rapidly. Suggested moves were also different.

No DC x64 version this time?

Best,

beachknight wrote:

Dann Corbit wrote:The new stockfish is really excellent against closed and semi-closed positions. It's very nice to have another option to analyze these positions.

I have tried both 1.7 and 1.7.1 in few closed midgame positions.
Almost all the time, evaluation value was bigger than what Rybka
or Naum or Shredder or other engine reports. And it changes very
rapidly. Suggested moves were also different.

No DC x64 version this time?

Best,

I would have to study the code for a while, because a lot has changed.

Well then you know what to do..this can be a new number one engine

Dann Corbit wrote: I would have to study the code for a while, because a lot has changed.

Yes, we did a major rewrite for search.

We will be waiting for your improvements

Ralph Stoesser wrote:From 1.6 to 1.7 we have seen a new search approach.
I wonder what comes next. A totally new eval?

This is great piece oft software. It plays already very strong. On the other hand there seem to be much potiential for further improvements. From reading the sources and the comments within the sources, I get the impression of a rough diamant here and there. Also it's eval output is a little tipsy some times. All in all to me it's still mysterious why it plays so strong???

Actually the eval is the same. The new thing (new for Stockfish) is to evaluate the position at every node and not only at the leaves as has been done until now.

This is new for Stockfish but it is a very known and old idea, for instance Glaurung 1 had the same concept and a lot of engines evaluate at each node. Normally this is slower, but in the new SF 1.7 it gives only a very little impact on ELO and allows to open the door for an whole new set of pruning techniques that rely on this...so at the end it ended up to be faster.

Here something for the next release!

http://sites.google.com/site/dshawul/home

Scorpio 2.5 Cluster beta!!

Time to make Stockfish 1.8 Cluster

Can be usefull information.

10 seconds per move. 1 CPU. O/P from STS Stat 2.5
http://sites.google.com/site/strategict ... st-results

Stockfish 1.7.1 replaced Critter as new #1 with 796/1000.

Dann Corbit wrote:

beachknight wrote:

Dann Corbit wrote:The new stockfish is really excellent against closed and semi-closed positions. It's very nice to have another option to analyze these positions.

I have tried both 1.7 and 1.7.1 in few closed midgame positions.
Almost all the time, evaluation value was bigger than what Rybka
or Naum or Shredder or other engine reports. And it changes very
rapidly. Suggested moves were also different.

No DC x64 version this time?

Best,

I would have to study the code for a while, because a lot has changed.

Just completed the first 1.7.1 version that will -probably, unless there are objections from Dann- have smooth nullmove as well as the "hijacked" Stockfish search. But if to be released then only after Rybka 4 I think : In the search I changed very little for now so not much point in publishing this.



The smooth reduction is not working yet in Rainbow Serpent, but the UCI options for it are already integrated in the Stockfish search and the engine is running, mostly with the Stockfish search still intact So much has changed in the search that there was not much point in merging any other changes before trying the standard search. Only 60 errors on the first try but that was really only one forgotten semicolon. Zugzwangdetection is off because I'm trying my idea of only storing the verification search results and then it may be less necessary. But build 001 is ready!

Eelco