TalkChess.com

yanquis1972 wrote:could someone explain to me (i ask in all ignorance/innocence) why stockfish doesnt do whatever it is asmfish does? i get something like 8MN/s w/ SF8 & over 10MN/s with asmfish. im using the bmi_2 compile, & its not explained by large pages, so is the speedup 'pure' or is there a performance hit as a result of it?

The number of nodes can be easily manipulated....
Dr.D

Dann Corbit wrote:

S.Taylor wrote:

stavros wrote:https://groups.google.com/forum/m/?_esc ... CoojE9O5jU

So, as long as Houdini hasn't got even much stronger yet again, then perhaps this new SF 8 is the clear best, and will still be competitive in the top ten, in ten years from now! (just like the old Rybka 4.1 is still in the top ten in some tests).

Rybka 4.1 is 5 years old, not 10.
5 years from now it will not be top 10.

But this komodo might make it for 10 years, in top ten. It seems harder to make stronger at the higher levels. [i didn't say that rybka kept in top 10 for 10 years yet].

Dr.Wael Deeb wrote:

yanquis1972 wrote:could someone explain to me (i ask in all ignorance/innocence) why stockfish doesnt do whatever it is asmfish does? i get something like 8MN/s w/ SF8 & over 10MN/s with asmfish. im using the bmi_2 compile, & its not explained by large pages, so is the speedup 'pure' or is there a performance hit as a result of it?

The number of nodes can be easily manipulated....
Dr.D

The strength increase is real.

Hi,

There seems to be no Stockfish 8 release note in the stockfishchess.org Website. Can somebody point me to it?

thanks

Jhoravi wrote:Hi,

There seems to be no Stockfish 8 release note in the stockfishchess.org Website. Can somebody point me to it?

thanks

https://stockfishchess.org/download/

gerold wrote:

Jhoravi wrote:Hi,

There seems to be no Stockfish 8 release note in the stockfishchess.org Website. Can somebody point me to it?

thanks

https://stockfishchess.org/download/

I think he was asking for the release notes, which I don't think have been updated yet. But all changes are documented on GitHub. Since Stockfish 7 was a long time ago, the number of changes is big. Mayor change is I think the SMP going from YBW with lots of branching points everywhere in the searchtree, to the a lot simpler 'Lazy SMP' where every thread is searching the same tree from the root up, and information exchange beteween threads is purely from hash. Because the threads don't have to exchange information about where in the tree they have to start searching, and no information about the node where they start, because they all start at the bottom, and there are no very short searches anymore starting deep in the tree, this is still a bit stronger in tests, although theoretically some contend it should be less efficient than translating alpha beta to parallel like YBW does. I don't quite remember, but I think that change was done already very soon after the release of Stockfish 7? So a long time ago... Many other changes, maybe someone can do a short synopsis. In the search for instance "Voyager One" made many changes to refine the use of the history tables, that helps also in one threaded searches.

Maybe Marco will document the number of lines added and removed, that should add up to quite a number. The total number of commits since Stock 7 will also be an impressive number.

SF switched to Lazy SMP on Oct 20 15, months before the Jan 2 2016 release of SF7.

Last winter I tried compiling a list of changes between SF6 and SF7, which resulted in the What's new in Stockfish 7 thread on chess.com. The list wasn't comprehensive or even quite accurate (e.g. some earlier changes - like Apex Pawns - were later discarded), but I think it provided a decent overview accessible to that part of the general crowd who wanted more info than what was in the short blurb on the official SF blog.

Perhaps one of these days I'll do the same for the SF7->SF8 changes, if someone better qualified or with more time on his hands doesn't beat me to it.

Right, thanks T. for correcting me there. I knew it was long ago but apparently SF7 is not yet a year old. It seems longer! I have not kept track of all the changes either. Wrote that from memory. But I do think there have been many changes everywhere, some don't seem big but it all adds up. Picking out a few does not do justice to all the work and the machine time that the testers put into it.