Honey XIr1 source released

Discussion of anything and everything relating to chess playing software and machines.

Moderators: Harvey Williamson, bob, hgm

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
MikeB
Posts: 4412
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Honey XIr1 source released

Post by MikeB » Wed Apr 01, 2020 5:03 am

https://github.com/MichaelB7/Stockfish/ ... /honey.zip

a modest 5 to 7 elo gain, but now with some of the HekaStockfish presearch code, modified, which I have labeled profound.

The profound (presearch) code does a much wider search early on - I offer two profound settings and they are both simply on and off switches.

As part of my testing , I had determined that the setting should be a higher value as the the time control increases this is done automatically by th e time control selected:

this setting is turned on by default for Honey BlueFish and Black-Diamond and off for Stockfish flavor XIr1 as that version always keep that in sync with cur-dev Stockfish functionally

option name Profound type check default true

the go infinite cand the go depth commands does not use a time control, so there is a second setting called :

option name Deep_Pro_Analysis type check default false which is always off by default.

It is to be used with 15 minute analysis . For longer analysis, say 1 hour , you might be better off using go movetime 3600000 ( or go mt 3600 , mt is in seconds . movetime is in milliseconds

You also now type 's' or 'set' to view all available command line terminal shortcuts
# Black-Diamond-XI-r1
Black Diamond XI-r1 by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
info string Hash LargePages 256 Mb
s

Shortcut Commands:

setoption name 'option name' value 'value'
is replaced by:
set (or 's'), 'option name' or 'option shortcut' 'value'

Note: set (or 's'), without an 'option' entered displays the shortcuts

'd' is the shortcut for 'depth'
'dpa' is the shortcut for 'Deep_Pro_Analysis'
'g' is the shortcut for 'go'
'i' is the shortcut for 'infinite'
'lp' is the shortcut for 'Large Pages'
'm' is the shortcut for 'Mate'
'mv' is the shortcut for 'MultiPV'
'mt' is the shortcut for 'Movetime'->

Note: 'mt' is in seconds, while
'Movetime' is in milliseconds

'p f' is the shortcut for 'position fen'
'pro' is the shortcut for 'Profound'
'sm' is the shortcut for 'SearchMoves'

Note: 'sm' or 'SearchMoves' MUST be the
last option on the command line!

't' is the shortcut for 'Threads'
'q' is the shortcut for 'quit'
'z' is the shortcut for 'SyzygyPath'
'?' is the shortcut for 'stop'
will create and start adding exe's to the release page tomorrow.
Image

Damir
Posts: 2427
Joined: Mon Feb 11, 2008 2:53 pm
Location: Denmark
Full name: Damir Desevac

Re: Honey XIr1 source released

Post by Damir » Wed Apr 01, 2020 9:59 am

Thanks Mike. Any executables available ?

User avatar
MikeB
Posts: 4412
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Honey XIr1 source released

Post by MikeB » Wed Apr 01, 2020 1:07 pm

Damir wrote:
Wed Apr 01, 2020 9:59 am
Thanks Mike. Any executables available ?
There is now!

https://github.com/MichaelB7/Stockfish/ ... /tag/Xi-r1

Windows, Mac Linux

Mac and Linux have officially dropped 32 bit support and so have I.
The base windows 32 bit and 64 bit versions are compiled for the lowest common denominator systems.
The Threadripper for Windows has bmi1 instructions , but not the bmi2 instructions ( and large pages is on by default)
Image

User avatar
Ovyron
Posts: 4399
Joined: Tue Jul 03, 2007 2:30 am

Re: Honey XIr1 source released

Post by Ovyron » Wed Apr 01, 2020 4:47 pm

Thanks Mike! It'll be interesting if with enough time the profound settings will leave humanity obsolete (since the positions where defaults don't find a move that profound does would be examples of ones where human interaction would have helped the engine play better moves. But now they're examples of how with profound human interaction isn't necessary for them.)

In my experiments I've seen instances where the engine doesn't find a move, not even with "presearch" at the start of the search, but where turning on "presearch" at a later point does find the move (it's just analyzing with SinglePV for some number of nodes and then switching to MultiPV at that point, and the best move is found), so I wonder if settings that make the engine switch between SinglePV and MultiPV at some points in the search will be the future (a future without scenarios where humans can help the engine...)

Dann Corbit
Posts: 11626
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Honey XIr1 source released

Post by Dann Corbit » Thu Apr 02, 2020 1:40 am

To Mike:
I notice in the downloads that you have a threadripper build.
I have found that (at least for long analysis periods) the builds with znver2 are not as fast as my ordinary modern builds.
Did you do some customization, or is it just Honey that benefits?
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

User avatar
MikeB
Posts: 4412
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Honey XIr1 source released

Post by MikeB » Thu Apr 02, 2020 2:44 am

Dann Corbit wrote:
Thu Apr 02, 2020 1:40 am
To Mike:
I notice in the downloads that you have a threadripper build.
I have found that (at least for long analysis periods) the builds with znver2 are not as fast as my ordinary modern builds.
Did you do some customization, or is it just Honey that benefits?
I'm not using znver1 or znver2 , did not mean to imply that but I understand why you asked that. I simply added bmi1 which can cause problems on machines which are truly just popcnt machines, hence the threadripper designation.
Image

MOBMAT
Posts: 327
Joined: Sat Feb 04, 2017 10:57 pm
Location: USA

Re: Honey XIr1 source released

Post by MOBMAT » Thu Apr 02, 2020 4:44 am

besides lots of hash, syzygy, and threads, what other setting maximize finding mates the quickest?
i7-6700K @ 4.00Ghz (using 6 threads), EGTBs on PCI SSD
Benchmark: Stockfish 11 64 bmi2 (nps): 2067669

peter
Posts: 2160
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: Honey XIr1 source released

Post by peter » Thu Apr 02, 2020 6:06 am

MikeB wrote:
Wed Apr 01, 2020 1:07 pm
There is now!

https://github.com/MichaelB7/Stockfish/ ... /tag/Xi-r1
Thanks a lot!
The modern compiles of all of of the engines can't use more than 16G hash at Win7x64.. At 32 they cause exception but only with more than 1 thread.
Nevermind, r1 ist tactically very strong.
Peter.

User avatar
MikeB
Posts: 4412
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Honey XIr1 source released

Post by MikeB » Thu Apr 02, 2020 11:50 pm

peter wrote:
Thu Apr 02, 2020 6:06 am
MikeB wrote:
Wed Apr 01, 2020 1:07 pm
There is now!

https://github.com/MichaelB7/Stockfish/ ... /tag/Xi-r1
Thanks a lot!
The modern compiles of all of of the engines can't use more than 16G hash at Win7x64.. At 32 they cause exception but only with more than 1 thread.
Nevermind, r1 ist tactically very strong.
Win 7 home or pro?
Image

kramnik
Posts: 36
Joined: Fri Mar 29, 2019 8:21 pm
Full name: Massimiliano Goi
Contact:

Re: Honey XIr1 source released

Post by kramnik » Fri Apr 03, 2020 3:36 am

Is there any comparison with a recent vanilla Stockfish in strength? I was thinking to put this engine on my website.

Post Reply