dchoman wrote:
I should probably add that while I have not tested it beyond 4 cores, I suspect the scaling beyond four cores is rather poor given my smp implementation. So I do not expect a big gain in playing strength from more.
- Dan
If you have the time and will, I can test 8 and 16 cores for you if you provide builds.
Best,
Martin
OK, no promises that they will work (I cannot test them here on my 4 core system), but I made 32 and 64 bit windows cross-compiles with any number of cores up to 16 possible. Just use the normal winboard "cores" command to change the number, the default is to use just 1 core.
dchoman wrote:
OK, no promises that they will work (I cannot test them here on my 4 core system), but I made 32 and 64 bit windows cross-compiles with any number of cores up to 16 possible. Just use the normal winboard "cores" command to change the number, the default is to use just 1 core.
When used from the command prompt, it works up to and including 16 cores when adjusting settings in the search.par file.
But when used with any amount of cores in analysis mode in Arena in the same way, it doesn't work. It starts over and over, the PVs go fast to depth 18 then starts over, then depth 19, starts over etc in a loop. And never use more than 1 core in the process.
dchoman wrote:
OK, no promises that they will work (I cannot test them here on my 4 core system), but I made 32 and 64 bit windows cross-compiles with any number of cores up to 16 possible. Just use the normal winboard "cores" command to change the number, the default is to use just 1 core.
When used from the command prompt, it works up to and including 16 cores when adjusting settings in the search.par file.
But when used with any amount of cores in analysis mode in Arena in the same way, it doesn't work. It starts over and over, the PVs go fast to depth 18 then starts over, then depth 19, starts over etc in a loop. And never use more than 1 core in the process.
Does it perhaps not support analysis mode?
Best,
Martin
Hi Martin,
Thanks for the test! In addition to the search.par file, you should also be able to issue the "cores" command from the command line interface as well, so "cores 4" would set it to use 4 cores from that point forward.
Regarding analysis in Arena, I am not certain as I have never tried it, but analysis works fine in winboard, xboard and scid ....
OK, I just loaded Arena, and I get exactly the same behavior as you describe in analysis mode. Old versions of EXchess behave exactly the same... like there is some limit on the search before it is reset. Not sure what the problem might be. I'll have to look into this, but it seems to play games just fine in Arena, as I have done that many times.
dchoman wrote:
OK, I just loaded Arena, and I get exactly the same behavior as you describe in analysis mode. Old versions of EXchess behave exactly the same... like there is some limit on the search before it is reset. Not sure what the problem might be. I'll have to look into this, but it seems to play games just fine in Arena, as I have done that many times.
- Dan
OK, I found the problem. EXchess does not implement the "." command during analysis. According the xboard protocol, the interface should stop sending "." after it receives no status update from the engine, but Arena keeps sending it periodically, which resets the search each time.
To properly implement the "." command during analysis is a bit of work as EXchess doesn't use a separate thread for I/O. If I can provide a work-around for this without too much trouble, I will issue an update... otherwise I will put it on the To-Do list for the next version.
Here is a report of the first 5 positions (I think the 1st position gave an error or something) from your testsuite, each run with 120 seconds, 16 cores and 4096 MB hash (it doesn't go any higher):
"WAC.001"; Smove: Qg6 Stime = 0.18 Right = 1/1 Total NPS = 135126 <sol.time> = 0.18
"WAC.002"; Smove: Rxb2 Stime = 0.17 Right = 2/2 Total NPS = 24549300 <sol.time> = 0.175 <depth> = 46 <time to depth> = 120
"WAC.003"; Smove: Rg3 Stime = 0 Right = 3/3 Total NPS = 21178724 <sol.time> = 0.117 <depth> = 43 <time to depth> = 120
"WAC.004"; Smove: Qxh7+ Stime = 0 Right = 4/4 Total NPS = 21177858 <sol.time> = 0.0875 <depth> = 43 <time to depth> = 120
"WAC.005"; Smove: Qc4+ Stime = 0 Right = 5/5 Total NPS = 21176982 <sol.time> = 0.07 <depth> = 43 <time to depth> = 120
Günther Höhne wrote:Have anyone a wb2uci for this engine?
Regards
Günther
I've found that using wb2uci plus a generic opening book for EXchess causes it to make its first move out of book almost instantly (if you recreate its own book file as an empty one).
Unless you're using the Fritz GUI, best to install as a Winboard engine.
Günther Höhne wrote:Have anyone a wb2uci for this engine?
Regards
Günther
I've found that using wb2uci plus a generic opening book for EXchess causes it to make its first move out of book almost instantly (if you recreate its own book file as an empty one).
Unless you're using the Fritz GUI, best to install as a Winboard engine.
Hello !
Try, please, this Wb2Uci.eng file. For me works perfectly !
==================
[ENGINE]
Name=EXchess 7.01b
Author=Daniel Homan, USA
Filename=Wb2Uci.exe
Wish you the best. Anxious to check it out.
