Crafty 22.0 Windows x64 available

[bob]

Hello Ernest,

I think infinite analysis, especially analyzing positions works only in single-cpu mode. I can't run epd tests in smp-mode either, as it hangs too! Sorry if I can't help you on this.

Denis

I did not notice that. Infinite analysis does not use smpnice at all since by definition, the engine should always be "thinking"...

[Gregory Owett]

"The current 22.0, with mt=2, ponder=off, no other crafty.rc file options will use both processors when thinking, and will use nothing when waiting on the opponent to move..."



Perfectly true, but it plays only one game, after he "exits"! Why?

It is not possible to make it play in a tournament.

[bob]

Do you have "mt=2" in your crafty.rc file? without that, crafty will only use one CPU...

So I have, with this in the Wb2Uci too and smpnice=0 in .rc task manager shows 100%, but I crash. Maybe I'll wait for another Windows executable but thanks again
regards
Peter.

At the moment it seems I've mananged it somehow. There was just too much hash maybe. To use both cores I have to have mt=2 not only in crafty.rc but also in Wb2UCI at command line of engine.
Great performance now
regards
Peter.[/quote]

Unfortunately, there is a bit of overlap between the crafty.rc/.craftyrc file and the GUI/winboard protocol. Here's the problem which is easy to understand once it is explained.

When you start crafty, the _first_ thing it does is to open the crafty.rc/.craftyrc file and read in and process all the commands included. Then it starts reading things from the GUI, and any commands coming from there override the crafty.rc settings since these new commands come after the .rc file has been processed.

Easy examples are things like "ponder=off". Which in text/console mode will disable pondering perfectly. But xboard is pretty set on sending an "easy" or a "hard" command depending on whether you use the -xponder option or not. And that will override the ponder=off from the .rc file.

So, bottom line, when something in your crafty.rc file appears to not be working, run winboard in the -debug mode or look at the log.nnn file to see what commands crafty is getting _after_ it has initialized and read the crafty.rc file. 99% of the problems are caused by this overlap. winboard has no mechanism to set the number of threads, but apparently the wb2uci program does. Whatever is sent to crafty _last_ is what will be used...

[bob]

"The current 22.0, with mt=2, ponder=off, no other crafty.rc file options will use both processors when thinking, and will use nothing when waiting on the opponent to move..."



Perfectly true, but it plays only one game, after he "exits"! Why?

It is not possible to make it play in a tournament.

I do not know what you mean? I ran a 5,120 game test last week using this and did not see a problem...

[Dr.Wael Deeb]

I downloaded Jim's Crafty build and using the crafty.rc which Werner was kind enough to send me,I finaly managed to use it as a MP chess engine and with ponder off when it's the opponent's time to think
This is under Arena 1.1

[Henrik Dinesen]

I downloaded Jim's Crafty build and using the crafty.rc which Werner was kind enough to send me,I finaly managed to use it as a MP chess engine and with ponder off when it's the opponent's time to think
This is under Arena 1.1

Wael, are you implying there has been any problems previosuly?

[Dr.Wael Deeb]

I downloaded Jim's Crafty build and using the crafty.rc which Werner was kind enough to send me,I finaly managed to use it as a MP chess engine and with ponder off when it's the opponent's time to think
This is under Arena 1.1

Wael, are you implying there has been any problems previosuly?

Definitely Henrik,if I manged to make it work as a MP chess engine,it did think during the opponent's time,and if I mange to make it ponder off during the opponent's time,it didn't work as a MP chess engine

[Gregory Owett]

Here are my settings :

Crafty.rc :

egtb
tbpath=h:\tbs
cache=32M
hash=96M
hashp=48M
mt=2
ponder=off
log=off
swindle=off
smpnice=0

If "smpnice=1" , Crafty crash immediatly.
With "smpnice=0", it works one game, but keeps 50% of memory, when it should not ponder.
Only with "mt=1" evrything works well, (but obviously, one proc used).

Wb2uci.eng :

[ENGINE]
Name=Crafty 22.0 PS64
Author=B.Hyatt
Filename=Wb2Uci.exe

[OPTIONS]
Program=crafty-22.0-win64.exe
WhiteScore=true
TerminateHard=true
RunIdle=true
ponder=off
log=off

When I do it play in a eng vs.eng match of several games, once the first game ended, it ceases to be active, leaving the GUI, (CB or Shredder), which informs to be awaiting Crafty, and the match is interrupted. Indeed Taskmgr confirms the absence of Crafty in the process.

[bob]

Here are my settings :

Crafty.rc :

egtb
tbpath=h:\tbs
cache=32M
hash=96M
hashp=48M
mt=2
ponder=off
log=off
swindle=off
smpnice=0

If "smpnice=1" , Crafty crash immediatly.
With "smpnice=0", it works one game, but keeps 50% of memory, when it should not ponder.
Only with "mt=1" evrything works well, (but obviously, one proc used).

Wb2uci.eng :

[ENGINE]
Name=Crafty 22.0 PS64
Author=B.Hyatt
Filename=Wb2Uci.exe

[OPTIONS]
Program=crafty-22.0-win64.exe
WhiteScore=true
TerminateHard=true
RunIdle=true
ponder=off
log=off

When I do it play in a eng vs.eng match of several games, once the first game ended, it ceases to be active, leaving the GUI, (CB or Shredder), which informs to be awaiting Crafty, and the match is interrupted. Indeed Taskmgr confirms the absence of Crafty in the process.

This is something both new and strange. I can recall any versions of Crafty in the past few years that would fail to play with mt=2. People have been using SMP crafty and winboard for years. I've been running SMP crafty using xboard and my own interface as well, and have not seen any SMP-related issues in any released version.

I have no idea what the problem is, except to say it can't be repeated on any of my unix boxes, nor on our team's windows boxes. It is possible a compiler is breaking something, or a GUI has a problem, or any of a dozen outside causes . But being unable to reproduce the problem makes it impossible to fix. If you could get rid of the log=off and send me the log files, they might show something although I doubt it...