Spike 1.4 is available

[bob]

Hi,

there is (afaik) no hash bug in Spike. The size is coming from a hudge sparsely filled mapping table used in eval. As it is sparsely filled there is no drawback for cache usage.
It´s not too hard to reduce the size, but it allows us to add special eval terms easier. We don´t expect a big drawback as today´s computer memory is much larger as the table itself.

Greetings Volker

Sparse is actually cache unfriendly. You can have one 32 bit value in a 64 byte block of memory, and you fetch the entire 64 byte block to get that 32 bit value. And you dislodge another 64 byte chunk in the process. And if you have lots of those, you use only 1/16th of cache, the rest being filled with data that is never accessed...

[Sylwy]

Hello Mr.Schäfer !
Hello Mr. Böhm !

Thank you a lot for your great chess engine !
Any bug observed !
Very good in my test suite.
Works well on my quad .
A good engine for analysis.



Thank you again !
Silvian

[PauloSoare]

Hi,

yes, we don´t reach stockfish strength. We had several years without much development and we are not on bitboards thus we don´t profit from 64bit processors.

Greetings Volker

I know the story of Spike. Volker, if you use bitboards I think you'll be among the best engines.

[Jouni]

BTW: which is the strongest non-bitboard engine at the moment?

Jouni

[Mangar]

Hi Jouni,

using bitboard will take a development of a full new engine. I like the idea of magic bitboards. But all that work

Greetings Volker

[PauloSoare]

BTW: which is the strongest non-bitboard engine at the moment?

Jouni

Maybe Schreder, Jouni?

[alpha123]

BTW: which is the strongest non-bitboard engine at the moment?

Jouni

Shredder.

Spike is catching up though.

Peter

[tomgdrums]

Hi all,

finally, after a long break of more than 4 years, a new version of Spike - 1.4 - is available!

There are more changes done in the last years than we can remember, so I want to give only a short summary:

- The evaluation part was first simplified, some terms with at least doubtable value were thrown out.
- The passed pawn evaluation was rewritten. Beside the common passed pawn evaluation, Spike has now some special terms and values for passed pawns in rook endgames and pawn endgames.
- Sort of an own material table. Chances are good, that the values of this table are sometimes by far off.
- Iirc a somewhat improved futility pruning.
- Much more selectivity by the extensive usage of lmr.
- Spike has now a strength reduction feature. Its not tested very extensively, so there are possibly some flaws.
- Support of up to 12 cores. Our own tests were only up to 6 cores, so we dont have any numbers regarding speedup above that.
- A new internal book with nearly 230.000 positions.
- Countless bug fixes (some of these bugs were really serious).

All in all we expect a large increase in playing strength, mainly resulting from a much deeper search, but the bugfixes and evaluation changes have their good share too!

Enjoy!

Best wishes
Volker and Ralf

http://www.spikechess.de/index_en.html

Hi guys!

Thanks for the great engine! It is fun to play against and analyze with. And thanks for the limit strength feature!

I wanted to let you know of a POSSIBLE mild bug. (it might be my fault!!) In the Shredder GUI (I can't attest to other GUI's) when I permanently set the engine to book learn and position learn it would crash if I tried to play two games in a row utilizing the limit strength feature. Once I set it to not use book or position learning then it ran fine when trying to play a few games in a row with the limit strength feature on.

Thanks! (i really do like Spike a lot!)

[Mangar]

Hi Tom,

thanks for the bug report. Have to check the book learn feature. I never tried book lerning with limit strength. IMHO this doesn´t really make sence.

Greetings Volker

[tomgdrums]

Hi Tom,

thanks for the bug report. Have to check the book learn feature. I never tried book lerning with limit strength. IMHO this doesn´t really make sence.

Greetings Volker

Hi Volker,

To make sure that I am not messing up....when I set the book and position learning as ON do I need to create a filename? And how should that filename read?

Thanks!