According to http://software.intel.com/sites/product ... _guide.pdf, page 9, Intel Nehalem has a 512 entry TLB and a 32 entry DTLB for large pages (2Mb/4Mb) per core. One can conclude that
- it doesn't matter whether threads or processes are used since each thread running on a separate core has its ...

1) Why does the memory access time depend on the amount of memory allocated?


TLB misses. Resolving a virtual address to a physical one can take up to 4 additional memory reads. Using large pages really helps a lot with larger hash sizes.

http://en.wikipedia.org/wiki/Translation_lookaside ...

1) Why does the memory access time depend on the amount of memory allocated?


TLB misses. Resolving a virtual address to a physical one can take up to 4 additional memory reads. Using large pages really helps a lot with larger hash sizes.

http://en.wikipedia.org/wiki/Translation_lookaside ...

Sorry for the confusion I unwillingly created.

Please read my post on talkchess:
http://www.talkchess.com/forum/viewtopic.php?t=47045

Hamfer wrote: What is the better data structure for performance?

Performance-wise they are equivalent. For convenience IMO it is best to use an union of both.

geko wrote:thanks Richard,
in first example, why
position p1 moves a2a3 a7a6 ???
should it be position p1 moves a2a3
a7a6 is the expected value

No. With UCI you never ponder the opponents move. You ponder your next move (as if the opponent already played the predicted reply).

as an example, is this a correct sequence of commands?
...
engine_1 -> bestmove a2a3 a7a6 ponderhit
engine_1 continues the search in ponder mode
engine_2 -> ponderhit
engine_1 plays a7a6, stops the ponder, continues the standard search and responds with bestmove b2b3
...

best regards
g ...

Terradon wrote:I do not know how other people think about it, but i just can't find all the pieces about chess programming all together at one place.

Hi,

Please take a look at this excellent site: http://chessprogramming.wikispaces.com/

Carlos Ylich wrote:For Richard Life
How do I clear the session file?

Among UCI options there should be a button named "Clear SF". Or simply delete the file while Critter is not running.

Javascript is quite slow no matter what you do, so I would suggest not going too fancy and stick with the "simpler = better" rule. Given that you do not intend to do any search, speed is not so essential. Just for generating legal moves I would go with the simple "int squares[64]" board ...