Ceibo 0.4 - new version

[federicorojo]

Hi all,
This is the latest release of Ceibo. I believe it is much better than the previous version.

I have published it here:
https://sites.google.com/site/ceibochess/

Or you can download it directly from here:
https://drive.google.com/open?id=1qL_0C ... bdCxbv5R0s


Thanks a lot !
Federico

[Graham Banks]

Thanks Federico.

[Sergio Martinez]

40/4 Update:

Code: Select all

Ceibo 0.4 64-bit	   2002	+22	−22	61.3%	−84.7	24.3%	777	 
Ceibo 0.3.65 64-bit	1706	+25	−25	51.9%	−10.2	21.2%	619
 

[federicorojo]

Amazing!!! I did not expect a ~300 elo improvement so this is great news.
@Sergio & Team, Thanks a lot for taking the time to test it.

I have a request though, I noticed that only 2 engines over 2000 ELO were matched against Ceibo, and my engine seems to beat both of them. Actually, Ceibo scored higher to all engines tested with except for one that would seem to be only slightly better than Ceibo. So it is a bit confusing for me. Just to be sure, I would like to see it lose against a couple more engines and not just the one.

That said, Would it be possible to test it against another 2 or 3 engines in the 2000 - 2050 range ?


Thanks again!
Fede

[Sergio Martinez]

Amazing!!! I did not expect a ~300 elo improvement so this is great news.
@Sergio & Team, Thanks a lot for taking the time to test it.

I have a request though, I noticed that only 2 engines over 2000 ELO were matched against Ceibo, and my engine seems to beat both of them. Actually, Ceibo scored higher to all engines tested with except for one that would seem to be only slightly better than Ceibo. So it is a bit confusing for me. Just to be sure, I would like to see it lose against a couple more engines and not just the one.

That said, Would it be possible to test it against another 2 or 3 engines in the 2000 - 2050 range ?


Thanks again!
Fede

You are welcome

Of course Fede.

We will continue testing your engine and in my case I will start a test against engines in that range of strength

Best regards

Sergio

[federicorojo]

Thank you !

[Sergio Martinez]

Thank you !

Some games more (opponents rated between 2000-2050):

Code: Select all

Ceibo 0.4 64-bit              :  228 (+ 89,= 47,- 92), 49.3 %

Oberon 0.04                   :  39 (+ 17,=  5,- 17), 50.0 %
Freyr 1.068 64-bit            :  38 (+ 17,=  8,- 13), 55.3 %
Zeta Dva 0303 64-bit          :  39 (+ 12,= 13,- 14), 47.4 %
Bestia 0.90                   :  39 (+ 17,=  7,- 15), 52.6 %
Bodo 0.2b                     :  39 (+ 11,= 10,- 18), 41.0 %
PLP 1661571                   :  34 (+ 15,=  4,- 15), 50.0 %

[federicorojo]

Thank you !

Some games more (opponents rated between 2000-2050):

Code: Select all

Ceibo 0.4 64-bit              :  228 (+ 89,= 47,- 92), 49.3 %

Oberon 0.04                   :  39 (+ 17,=  5,- 17), 50.0 %
Freyr 1.068 64-bit            :  38 (+ 17,=  8,- 13), 55.3 %
Zeta Dva 0303 64-bit          :  39 (+ 12,= 13,- 14), 47.4 %
Bestia 0.90                   :  39 (+ 17,=  7,- 15), 52.6 %
Bodo 0.2b                     :  39 (+ 11,= 10,- 18), 41.0 %
PLP 1661571                   :  34 (+ 15,=  4,- 15), 50.0 %

Awesome! it is much appreciated.

Thank you for taking the time to do this.

[Sergio Martinez]

Thank you !

Some games more (opponents rated between 2000-2050):

Code: Select all

Ceibo 0.4 64-bit              :  228 (+ 89,= 47,- 92), 49.3 %

Oberon 0.04                   :  39 (+ 17,=  5,- 17), 50.0 %
Freyr 1.068 64-bit            :  38 (+ 17,=  8,- 13), 55.3 %
Zeta Dva 0303 64-bit          :  39 (+ 12,= 13,- 14), 47.4 %
Bestia 0.90                   :  39 (+ 17,=  7,- 15), 52.6 %
Bodo 0.2b                     :  39 (+ 11,= 10,- 18), 41.0 %
PLP 1661571                   :  34 (+ 15,=  4,- 15), 50.0 %

Awesome! it is much appreciated.

Thank you for taking the time to do this.

You are welcome

[T-Poppins]

I have two issues to report with this version. Cannot find any contact info for Federico, so I'm posting it here.

1) Ceibo 0.4 crashes instantly without making any move on my main desktop.

Dual Xeon X5670 @2.93 GHz, 12 cores, HT off, 40 GB RAM
Windows 7 x64 Pro
Cute Chess GUI v1.1.0
512 MB hash

Crash mini-dump with two associated text files:
http://www.mediafire.com/file/obbkqzsu7 ... -crash.zip

Oddly enough, no crashes on my dual E5-2670, same OS, same GUI, same settings.

2) Random 0.00 evals.

Ceibo usually sends normal UCI output to the GUI (info score cp etc.) and then suddenly the last "info score" line will have cp 0, even in clearly won/lost positions. The lines before the last one are fine, it's always the last one (the one that is saved to PGN) that is the problem.

This typically affects 30-60% of the moves in a game. Not only it looks senseless, it also interferes with automatic game adjudication.

[T-Poppins]

Here's the log of communications between the GUI and Ceibo:

Code: Select all

0.000: - Starting server: Ceibo_v0.4.exe
0.000: - Priority class set to normal
0.000: - Thread priority set to normal
0.000: > uci
0.015: < Ceibo
0.015: < id name Ceibo v0.4
0.015: < id author Federico Rojo
0.015: < option name Hash type spin min 1 max 512 default 256
0.015: < option name NullMove type check default true
0.015: < uciok
0.015: > setoption name Hash value 256
0.015: < Hash Table created with 16777215 records ( 256 mb )
0.015: > setoption name NullMove value true
0.015: > isready
0.078: < readyok
0.078: > ucinewgame
0.078: > position startpos
0.078: > isready
0.109: < readyok
0.109: > go wtime 1260000 btime 1260000 movestogo 40
13.478: - Server closed
13.478: - Stopping server communication
13.509: - Closing log

The game is terminated as "abandoned", with a "White disconnects" message.
Default settings, Ceibo white, no book, 40/21 TC (the equivalent of CCRL's 40/40 for this box).
I'll reboot and see if that helps. Just checked my other test box, a dual E5-2690v2, and no crashes there.

[federicorojo]

Hi Tirsa,

Thanks for taking the time to test Ceibo. In regards to the issues:

#1. Seems quite odd. It would appear it crashes between calculating the amount of time to allocate to that move and right before the main search function is triggered. I cant seem to find anything that could cause that exception. I ll have to do a bit more digging. Could you check if you have the same problem with the previous version 0.3.65 ?

#2. I had noticed that issued before but never got around to fix it. I ll work on it.


Thanks again! Fede

[T-Poppins]

Just tried v0.3.65 at its default settings -- same crashes. When used from console rather than the GUI it crashes right after ucinewgame, unlike v0.4 which lets me as far as go depth 10.



Here's an excerpt from the XML accompanying the crash dump:

Code: Select all

	<ProblemSignatures>
		<EventType>APPCRASH</EventType>
		<Parameter0>Ceibo_v0.3.65.exe</Parameter0>
		<Parameter1>0.0.0.0</Parameter1>
		<Parameter2>5a334356</Parameter2>
		<Parameter3>Ceibo_v0.3.65.exe</Parameter3>
		<Parameter4>0.0.0.0</Parameter4>
		<Parameter5>5a334356</Parameter5>
		<Parameter6>c000001d</Parameter6>
		<Parameter7>0000000000003f8f</Parameter7>
	</ProblemSignatures>
	<DynamicSignatures>
		<Parameter1>6.1.7601.2.1.0.256.48</Parameter1>
		<Parameter2>1033</Parameter2>
		<Parameter22>a4d9</Parameter22>
		<Parameter23>a4d9b8a3a652487d43c1846f7d5c6ed6</Parameter23>
		<Parameter24>2b50</Parameter24>
		<Parameter25>2b5011b211de4b34dd6ba88b8c71280a</Parameter25>
	</DynamicSignatures>

A reboot didn't help. Bizarre!
Thank you for looking into it, Fede.

[federicorojo]

ok, here is a stretch... have you installed service pack 1 on your win 7 ?

[T-Poppins]

Yes. Same OS and updates installed on all three boxes.

This one is my general-use desktop, so it has quite a few more processes running in the background than the other two. I tried shutting down all background processes and services one by one until I thought I was close to the bare minimum, with no improvement. I'll try booting into Safe Mode next.

Just noticed another issue. Kind of related to #2. When Ceibo is about to be mated it shows "score mate x" (positive value), whereas it should be a negative value. That also makes auto-adjudication nearly impossible.

From UCI protocol specs: * score * cp the score from the engine's point of view in centipawns. * mate mate in y moves, not plies. If the engine is getting mated use negativ values for y.

[T-Poppins]

I booted into Safe Mode, same results. Oddly, after updating the ESET virus scanner/firewall and another reboot Ceibo stopped crashing. Now it merely "disconnects" without making a single move, and the games are terminated as "abandoned".

No worries, I have the Ceibo 40/40 gauntlet nearly complete on my 16-core test box (same OS, same ESET), so expect to see your engine on the 40/40 list by Sunday.

[federicorojo]

Amazing!. Thanks a lot.

As for the issue specifically with your Dual Xeon X5670, it makes me think that perhaps your windows might be corrupted. Perhaps a fresh install could fix it.
In any case, I ll be sure to have the other issues you mentioned fixed by my next release. Thanks again, and looking forward to seeing those results!


Fede

[T-Poppins]

Tested v0.4.1, still crashes. I doubt the problem is with this system, as I test hundreds of engines on it and get most to work, except the most buggy ones.

Could it be the compiler's settings? If you use aggressive optimization settings like -O3, perhaps it would be worth it to try lowering it? Or if it's a POPCNT compile, making a plain 64-bit one.

BTW, what's new in version 0.4.1?

[federicorojo]

Hi Tirsa,

I use the ms compiler that comes with the Visual Studio 2015, and not a gnu compiler. Also, I compile it with pgo, so the compiler chooses the best optimization flags. That said, I still believe there may be some issue with that particular computer, as I don't use any special flags to target any particular hardware, except x64 and therefore should work just fine on any x64 computer running windows.

In any case, to troubleshoot the issue, I have attached different versions so that you may try out. The different variants have different filenames which include forcing AVX , SSE (these are hardware specific , and not generic as my regular releases), and one more compiled without profile guided optimization. 2 of these versions run slower on my machine than the regular no target specific releases. My gut feeling says that the AVX version would work for you.

Let me know if any work for you or if more than one !


https://drive.google.com/open?id=1QudVq ... 34JlswRHXF


Thanks, Fede

[T-Poppins]

Thank you for taking the time look into this, Fede.

There seems to be a misconfiguration in your setup, or else a bug in VS 2015. Every one of the three builds you posted crashes with the same exception error, e.g.

Code: Select all

Faulting application name: Ceibo_0.4.1_noPGO.exe, version: 0.0.0.0, time stamp: 0x5bd46c2e
Faulting module name: Ceibo_0.4.1_noPGO.exe, version: 0.0.0.0, time stamp: 0x5bd46c2e
Exception code: 0xc000001d
Fault offset: 0x00000000000055d8

The offset varies between the three, the code is the same.

The "uci - isready - ucinewgame - position startpos" sequence is as far as I can get, the "go" command (e.g. "go depth 20" or "go infinite") causes an instant crash.

Exception code 0xc000001d is indicative of a CPU failing to decode an instruction (note that the originator of the linked Stackoverflow uses VS 2015 as well). The X5680 Xeon is a Westmere-EP chip, which means support for SSE up to v4.2. There is no way it would fail run a proper POPCNT build, much less a vanilla 64-bit non-PGO one.

Currently I have over 460 engines (including subversions) installed and tested on this box. Additionally, I have a list of 23 -- all but one under 1800 Elo, most older than 10 years -- tested and found too buggy for regular use. And then there are just two that even in console mode refuse to run outright on this box and crash as soon as you try to execute any command. One of them is Napoleon 1.8 which only has BMI2 bins released. The other is Ceibo.

My other two testing dual Xeon rigs - a Sandy Bridge and an Ivy Bridge -- have no problems with Ceibo. The main difference between them and this chip seems to AVX support. That's the clue to the mystery, I believe.

The Sandy Bridge box is in the storage for the time being, which makes me more interested in being able to run Ceibo on this Westmere-EP. Sure, with 800 games on the 40/40 list Ceibo is looking good now; there is no reason why it shouldn't be used as a worthy opponent for other engines, though, now is there?

[T-Poppins]

There's also the question of the latest v0.4.1 release. Should it be merged with v0.4 on our lists as a mere bug fix, or does it have enough changes to necessitate testing it as a separate version? If you think the differences may be worth at least 20 Elo we should be able to corroborate it.

[Ray]

The exe in the original post does not work on my AMD Phenom II X6 either. That is, at the "Go" stage it crashes.

And neither do the 3 exes above. The AVX one does execute and initially respond to the UCI command, but it should not execute at all, because the CPU has no AVX support.

All of them are OK on my AMD FX8350 which does have AVX support.

Definitely something wrong with the compiles and not Tirsa's machine. And as Tirsa says, it is probably AVX related.

[T-Poppins]

The Phenom II supports POPCNT but not AVX, which seems to confirm my hypothesis.

[Ray]

The Phenom II supports POPCNT but not AVX, which seems to confirm my hypothesis.

Yes, see my edited post.

[federicorojo]

There's also the question of the latest v0.4.1 release. Should it be merged with v0.4 on our lists as a mere bug fix, or does it have enough changes to necessitate testing it as a separate version? If you think the differences may be worth at least 20 Elo we should be able to corroborate it.

Version 0.4.1 only addresses the 2 issues you had reported. 1. When the engine is being mated did not include negative values. 2. Last line including zero scores.

There is no other inprovement there.