Crafty 22.0 Windows x64 available

[Dirt]

Hi Jim,
no!
The 32bit engine crashes too when using smpnice=1

I had it run a short match overnight and the 32 bit version crashed on the 17th game on my single cpu system. My .rc file:

Code: Select all

hashp=16M
log=off
tbpath d:\tb
egtb
cache=16M
ponder off
swindle off
timebook 400 2

[Dr.Wael Deeb]

Hi Jim,
no!
The 32bit engine crashes too when using smpnice=1

I had it run a short match overnight and the 32 bit version crashed on the 17th game on my single cpu system. My .rc file:

Code: Select all

hashp=16M
log=off
tbpath d:\tb
egtb
cache=16M
ponder off
swindle off
timebook 400 2

Nothing wrong with your .rc file,it seems that something is broken inside the source code

[Jim Ablett]

Hi Jim,
no!
The 32bit engine crashes too when using smpnice=1

I had it run a short match overnight and the 32 bit version crashed on the 17th game on my single cpu system. My .rc file:

Code: Select all

hashp=16M
log=off
tbpath d:\tb
egtb
cache=16M
ponder off
swindle off
timebook 400 2

Hi Greg,

I think all the big changes to the src code in this version
have introduced bugs and instability - I only taking about Windows here,
I don't know how it performs on Linux.

I've tested 32 bit build SMP on my Core2Duo (2 threads) and it's working ok
and I've just re-compiled x64 and SMP is working ok again (2 threads).
Download from my homepage.

regards,
Jim.

[Werner]

Hi Jim,
I just started an eng - eng match under arena with smpnice=1 in rc-file - now using Intel Core2duo processor.
I have the same behaviour as before. Crafty does not move. The engine crashes and is no more in memory.
As I told before : when not using this new command, the engine is running ok here, but it uses 1 CPU when its idle.
I think this is not your fault - its something wrong with the code and only Bob can change it.

regards

[peter]

Thanks a lot, with this compile x64 runs fine on chessbase GUI and dual core.
But how can I be sure both cores are active? Crafty shows about 1300 to 1500kN/s
regards
Peter.

[bob]

Hi Werner,

I have 2 questions:
a) does it work for you with that rc file? Why 2 lines with egtb?

I don't know why the second one is there

b) do you remember the line which stops using 1 CPU during the other engine is running?

Wael has kindly answered that one. I didn't know.

smpnice=1

smpmt=2

is the one which may not work at all.

regards,
Jim.

Either one works. mt=2 is something I had planned on phasing out, so that all SMP commands would start with "smp" for clarity. smpmt=2 and mt=2 do exactly the same thing. I did forget to make "smpnice=1" the default, which I just did in my source which will become the standard for 22.1. BTW, do not forget that with ponder=on, smpnice=1 means nothing. It only works in ponder=off games. And there is a cost for using it so it should not be used in games at very fast time controls like 1+0 and the like, because doing a fork() before each search is not free in terms of overhead...

There is one thing I forgot to do in 22.0 which I now regret, but may well fix anyway, I now have an ugly way of initializing the random numbers that was done so that the old book.bin/etc files continue to work. It is not a very cache-friendly way of doing things, although the arrays are small and it is only done once at the start of execution. I had planned on simply requiring that 22.0 have a new book.bin as the random numbers will change a bit, but I totally forgot about it with all the testing going on.

[bob]

Hi Jim,
I tried to run it with 2 CPUs under arena and started a eng-eng match.
Crafty startes running on 2 CPUs a few seconds and then crashes before first move after book is made. This test was made on my AMD 64x2.
Here is the rc file I use:
egtb
hash=512M
hashp=32M
log=off
tbpath=j:\tbs
cache=16M
mode=tournament
swindle off # (Allows Crafty to try to win drawn games (according to Endgame Tables))
mt=2 #(Increases Crafty's MaxThreads to 2 for a dual CPU computer)
timebook 80 8
#adaptive NPS a
smpnice=1

edit:
just using it without smpnice=1
match is running - crafty uses 1 CPU when other engine is running

First, I assume this is ponder=off? I will test it again as it is possible I broke something from 21.7 where this was added, to 22.0 which is current... In fact, it appears that I did. I am looking at it which means that a new 22.0 will be forthcoming. Tracy and I were trading version changes back and forth so frequently in the CCT-preparation, that we lost several changes when versions crossed in the mail. Something is broken here...

OK, I was correct. A change to thread.c was lost in the melee before the last CCT event. I will have a new 22.0 source up in 5 minutes or so, should fix this not terminating the threads properly problem, and also will have smpnice=1 as the default...

BTW when you download this version, please delete _all_ old .c and .h files. A couple of .c files are no longer needed in this new version and can cause some confusion if you try to use them with an old Makefile...

Source is now up...

[bob]

Hi Jim,
I tried to run it with 2 CPUs under arena and started a eng-eng match.
Crafty startes running on 2 CPUs a few seconds and then crashes before first move after book is made. This test was made on my AMD 64x2.
Here is the rc file I use:
egtb
hash=512M
hashp=32M
log=off
tbpath=j:\tbs
cache=16M
mode=tournament
swindle off # (Allows Crafty to try to win drawn games (according to Endgame Tables))
mt=2 #(Increases Crafty's MaxThreads to 2 for a dual CPU computer)
timebook 80 8
#adaptive NPS a
smpnice=1

edit:
just using it without smpnice=1
match is running - crafty uses 1 CPU when other engine is running

Hi Werner,

Yes, there is a problem with SMP in x64.

I just tested the 32 bit again in smp mode (2 core) and it's
seems to be working ok.

Jim.

This one is unknown to me. I always run in 64 bit mode on my core-2 and it is working fine for me using either gcc or icc (Intel's compiler). I'm not aware of any SMP issues at all on any platform we have tested on. Peter and Mike run windows tests for us all the time, 32 and 64 bit, and they are also not reporting any problems with the current version... I just ran an 8cpu test and it also ran perfectly under 64 bit linux. No idea what is happening there unless there is a compiler/library problem...

[bob]

Hi Jim,
I just started an eng - eng match under arena with smpnice=1 in rc-file - now using Intel Core2duo processor.
I have the same behaviour as before. Crafty does not move. The engine crashes and is no more in memory.
As I told before : when not using this new command, the engine is running ok here, but it uses 1 CPU when its idle.
I think this is not your fault - its something wrong with the code and only Bob can change it.

regards

There was an issue with smpnice=1. I had fixed it, but the fix was lost during our CCT preparations. I just uploaded a new source package (22.0 still) that works with smpnice=1 on my 2 and 8 core boxes. The old one would hang and/or crash with smpnice=1...

I am not aware of any other problems or instabilities in this version, however. During the last 2 days I have played over 50,000 games on our cluster with absolutely zero problems, against 5 different opponents... Some with 1 processor, some using 2.

[bob]

Thanks a lot, with this compile x64 runs fine on chessbase GUI and dual core.
But how can I be sure both cores are active? Crafty shows about 1300 to 1500kN/s
regards
Peter.

click on the task bar at the bottom of the screen and then click on task manager and then performance (I think) to see if it thinks both cpus are running.

NPS should double with 2 cores over 1.