View previous topic View next topic Go down
Author Message
Angels77
Admin
Admin
Posts: 457
Reputation: 1164
Join date: 2009-11-24
Location: not where yoraj thinks
PostSubject: Chessultant - UCI chess-cluster-engine for Ippolit ? Today at 10:19 Select/Unselect multi-quote Reply with quote Edit/Delete this post Delete this post View IP address of poster
Chessultant - UCI chess-cluster-engine
1 - 1 of 1
Angels_77 Angels_77 3 minutes ago
Chessultant maker (Dr.) Alexander Schmidt.posts
I already regretted wasting my time with the Rybka's (not tolerating the current best UCI-engines out there: Fire(Bird) & IvanHoe), so i hopefully can be a co-worker of the decembrists soon,
Viva la revolution and Best wishes,
Alex
you can run 4x Robbolito (you could even run up to 65536 ) on a
single quad-core, if engines don't support multi-instances natively copying&
renaming them often does the job (which does Chessultant automatically,
it also intelligently attaches them to cores), there is also a VMWare-option.
And yes, it's already possible to connect to other computers over the internet,
i had a few nice tests in the last week. Of course, the persistant-hash-option
(which moves the parents or best-suited persistant-hash-file to the slave-
node) in most cases doesn't make to much sense over the internet because
uploading MB's or even GB's takes its time, and of course the slaves should
have descent speed and being pretty symmetric (homogen) is always an
advantage, i'm currently working on algorithms for bigger tolerance&usability
on non-symmetric (heterogen) clusters... you will be able to choose a
symmetric (less logic required=quicker) and non-symmetric-mode.
i also see one engine per core (or per chip if you have enough computer-
slaves available) making the most sense from a performance point of view,
but it all depends on what modes or custom search-patterns you want to use.
Yes, a beta-release to technically interested and helpful hands is planed
which will hopefully start in may (=in the next few weeks). It was originally
planed with 16 people in mind when i started the original thread in the rybka-
forum, but while finding a lot more was easy only a few of those seem to be
willing to improve the project as itself, and to those i will restrict the beta,
As you may have followed i'm also in talks for a "Fire edition" or "IvanHoe edition".
I'm currently working at several things including a database-concept and -scheme
to save the deep analysis which was already processed by Chessultant before,
creating modes and custom search patterns intended for a wider (non-huge-
cluster-)audience and unsymmetrical (heterogen) architectures&systems,
and improving here and there, especially the little Qt-based GUI which comes
optionally with it (to set the parameters and custom search patterns, giving
a -to my taste- pretty nice visual control of the whole calculation&processing
and also a look into the evaluations which can't be covered by the UCI-standard
and therefore wouldn't make it into your favorite UCI-compatible GUI and end up
just in the loggin-files). I'm also creating a little (startup-)documentation&help,
so you hopefully will accept and like it.
Btw., there are no bugs
Viva la revolution and Best wishes,
Alex
- Angels_77 Angels_77
we say
sign us up as Beta testers please Doc
Viva the Revolution Commrades
http://ippolit.wikispaces.com/message/v ... F/37017446
permalink
http://morethanchessagame.forumotion.co ... polit#4444
_________________
nothing is ever truly lost , just missplaced and awaiting us