silly question about board representation

[krismchess]

I have couple of very basic & silly question about board representation:

1) Does the choice of board representation affect the program strength (ELO rating) in general? I am looking for 0x88 vs bitboard approach. I know bitboard approach is modern and fast - but does that affect the strength of the chess engine. Are there any research done in this topic?

2) Does the search depth gets affected because of the choice of board representation? Again I want to compare between 0x88 vs bitboard approach.

Thanks,
Kalyan

[AndrewGrant]

It's really up to you. There are engines that use both ways which are very strong.

I, however, tend to believe that bitboards are better, do to the ability to evaluate boards using bitboards quickly. To each his own.

Speed certainly plays a factor, but not as much as you might expect. The numbers I see floating around say that if you double the speed of your engine, i'm raw NPS (not changing search algorithm or evaluation algorithm) then you'll see a gain of 50-70 ElO

[CheckersGuy]

It's really up to you. There are engines that use both ways which are very strong.

I, however, tend to believe that bitboards are better, do to the ability to evaluate boards using bitboards quickly. To each his own.

Speed certainly plays a factor, but not as much as you might expect. The numbers I see floating around say that if you double the speed of your engine, i'm raw NPS (not changing search algorithm or evaluation algorithm) then you'll see a gain of 50-70 ElO

That actually makes me wonder, how much faster engines, that are written in C, are compared to, for example, an engine written in Java.

If someone replicated stockfish using Java, how much worse weould that engine actually be ? 100-150 elo ? No one would ever do this but I would still like to know

[Sven]

1) Does the choice of board representation affect the program strength (ELO rating) in general? I am looking for 0x88 vs bitboard approach. I know bitboard approach is modern and fast - but does that affect the strength of the chess engine. Are there any research done in this topic?

Yes, but not much.

2) Does the search depth gets affected because of the choice of board representation? Again I want to compare between 0x88 vs bitboard approach.

Same here: yes, but not much

Dominant factors for chess engine playing strength are:

- absence of bugs;
- a strong search, implemented correctly, that sees the most important lines of play with the smallest possible tree size;
- a well-tuned evaluation function;
- an excellent testing environment.

Speed helps but is not the most decisive part.

[theturk1234]

Honestly I've found that bitboards are much better than 0x88 rep.
Their basic advantage is that they can be used much easier than other representations. For example, the occupancy is literally one instruction.

If you want to see some expert bitboard usage, check the Stockfish source code.

David Cimbalista