1 core can be better than 6 cores (Stockfish)

[Jouni]

With latest Stockfish I got again: 1 core 175 but 6 cores 171. But note, that total used time is less with 6 cores. So it's better. In testsuites used time is more important than solved positions! BTW SF17 solves 185 with 6 cores .

[Paloma]

This suite is simply too easy, (many older items 2008-2011,
the current engines solve them in fractions of a second) which increases the fluctuations.
Maybe we should integrate the 20-30 unsolved tasks into a better suite and rename it.

[Werewolf]

This suite is simply too easy, (many older items 2008-2011,
the current engines solve them in fractions of a second) which increases the fluctuations.
Maybe we should integrate the 20-30 unsolved tasks into a better suite and rename it.

Yes.
The suite is both too easy and has cooked positions. It is impossible to deduce anything from it at the moment.
I'm trying to clean the suite up and incorporate Hai's testsuite.
This is slow as to be thorough every legal move needs consideration. I've done about 30 positions so far.

[Werewolf]

Well that was a labour.

I went through both testsuites and distilled them, removing:
a) All positons that are cooked. Often the "best" move was just a pretty move and the engine cannot be blamed if it finds an ugly win instead.

b) All positions which have nothing to do with tactics and therefore potentially no impact on hardware. Usually these are positions which are EF dependant rather than search/calculation dependant, like understanding a cut off piece is out of the game for good or that a certain 7 man tablebase combo is winning etc. Hardware will not affect the evaluation or move choice here.

c) All positions which are too easy and solved in a second or less.

This led to only 44 positons salvaged from Hai's 115 testsuite and 33 positions from the Arasan 200 testsuite. Given there are probably some duplicates in both, this leaves only around 60 positions in total that are high quality. It's getting tougher and tougher to build a good testsuite.

[Paloma]

I think there is an issue with the Arasan Testsuite, unfortunately. Some positions are so easy they are found well under a second making it impossible to distingush between 1 core, 8 cores or 512 cores etc etc.

Some positions are cooked. For example, number 43 has an alternative move which is a fraction of a pawn less good than the main solution, but still very good.

I'm now moving on to Hai's suggestion.

number 43
My further run with longer time resulted in:
1...e4 -3.50 at depth 51
2nd best :
1...Nh2 -2.71 at depth 51
(but with old SF 11)

so 1...e4 is the best move

[Werewolf]

I think there is an issue with the Arasan Testsuite, unfortunately. Some positions are so easy they are found well under a second making it impossible to distingush between 1 core, 8 cores or 512 cores etc etc.

Some positions are cooked. For example, number 43 has an alternative move which is a fraction of a pawn less good than the main solution, but still very good.

I'm now moving on to Hai's suggestion.

number 43
My further run with longer time resulted in:
1...e4 -3.50 at depth 51
2nd best :
1...Nh2 -2.71 at depth 51
(but with old SF 11)

so 1...e4 is the best move

Yes but it's less than 1 pawn difference, if the engine finds ...Nh2 in a second it can't really be blamed in a 15 second testsuite - it still wins

[Vinvin]

This led to only 44 positons salvaged from Hai's 115 testsuite and 33 positions from the Arasan 200 testsuite. Given there are probably some duplicates in both, this leaves only around 60 positions in total that are high quality. It's getting tougher and tougher to build a good testsuite.

Hi !
I'm very interested in getting this selection of about 60 positions !
Please copy/paste the EPDs here.
Thanks,
Vincent

[Paloma]

Hi,
I'm also interested.

Thanks

[Werewolf]

This led to only 44 positons salvaged from Hai's 115 testsuite and 33 positions from the Arasan 200 testsuite. Given there are probably some duplicates in both, this leaves only around 60 positions in total that are high quality. It's getting tougher and tougher to build a good testsuite.

Hi !
I'm very interested in getting this selection of about 60 positions !
Please copy/paste the EPDs here.
Thanks,
Vincent

I've saved it as a CBH database, how do I get the epds from that?

[Paloma]

I don't think that works, but it should work as a *.pgn file. Create a new database (select file type pgn on the bottom), copy all games from cbh game list (ctlr a, ctrl c), then paste it into pgn file.

But maybe Vincent knows more about it.

Thanks