Unofficial World Championship Of Computer Programs In International Draughts 2021

[Krzysztof Grzelak]

I cordially invite you to the tournament " The Unofficial World Championship Of Computer Programs In International Draughts 2021 ". For all information about the tournament, please visit https://www.uwccpi2021.draughtsprograms.eu/.

For now, individual programs will participate in the tournament:

Ares - Joost Buijs - Netherlands
Aurora Borealis Professional - Alexander Svirin - Russia
Cerberus - Leo Nagels - Netherlands
Damage - Bert Tuyt - Netherlands
Damira - Alexander Osipov - Belarus
Dragon - Michel Grimminck - Netherlands
Dream - Saïd Koudache - France
Edeon Sport - Alexey Odnoklubov, Vasily Naumik, Victor Naumik - Belarus
Flits - Adri Vermeulen - Netherlands
GWD - Gijsbert Wiesenekker - Netherlands
Kingsrow - Edgar Gilbert - United States of America
Maximus - Jan-Jaap van Horssen - Netherlands
Moby Dam - Harm Jetten - Netherlands
Plus 500 - Serge Startsev - Russia
Scan - Fabien Letouzey - France
Sjende Blyn - Jelle Wiersma - Netherlands
TD King - Ton Tillemans - Switzerland
Tornado - Frank Mesander - Netherlands
Truus voor Windows - Stef Keetman - Netherlands

Thanks to everyone I mentioned on the website. I would like to add that individual versions of programs may change. Please also let developers make their programs available a week before the tournament. I also invite other programs to participate in the tournament.

[Krzysztof Grzelak]

Who can from programmers to make the program available for the tournament - please. Many thanks in advance.

[Joost Buijs]

Who can from programmers to make the program available for the tournament - please. Many thanks in advance.

Krzysztof,

I will send you my program within a week or so, let's say after next weekend on Monday the 9th of Augustus. Unfortunately I didn't have time to add pondering and to make a book, but it will play a decent game without.

I'm not sure if it will work on your AMD-TR1 without modification, if it doesn't it will not not be too difficult to adapt.

It has no GUI whatsoever, it can only run as DXP server. I've checked it with the Dam 2.2 v7 GUI and it seems to work. When you think of another solution let me know beforehand.

Joost

[Krzysztof Grzelak]

Krzysztof,

I will send you my program within a week or so, let's say after next weekend on Monday the 9th of Augustus. Unfortunately I didn't have time to add pondering and to make a book, but it will play a decent game without.

I'm not sure if it will work on your AMD-TR1 without modification, if it doesn't it will not not be too difficult to adapt.

It has no GUI whatsoever, it can only run as DXP server. I've checked it with the Dam 2.2 v7 GUI and it seems to work. When you think of another solution let me know beforehand.

Joost

At the very beginning, thank you very much Joost. Too bad you didn't make an option pondering. Nevertheless, thank you very much Joost.

[Joost Buijs]

Krzysztof,

I will send you my program within a week or so, let's say after next weekend on Monday the 9th of Augustus. Unfortunately I didn't have time to add pondering and to make a book, but it will play a decent game without.

I'm not sure if it will work on your AMD-TR1 without modification, if it doesn't it will not not be too difficult to adapt.

It has no GUI whatsoever, it can only run as DXP server. I've checked it with the Dam 2.2 v7 GUI and it seems to work. When you think of another solution let me know beforehand.

Joost

At the very beginning, thank you very much Joost. Too bad you didn't make an option pondering. Nevertheless, thank you very much Joost.

Krzysztof,

The tournament will be in September, maybe there is still time to add pondering after vacation. First I will send you a copy without.
This week I will put a preliminary version of the program on Dropbox, this gives you the opportunity to test it on your AMD-TR1.

I wonder what kind of speed it will reach on the TR1, on my Intel machine (using 16 cores with AVX2) it does about 55 Mn/s, this is about 3 to 4 times as slow as what the pattern version does. I expect the TR1 to reach something like 35 Mn/s.

Joost

[Krzysztof Grzelak]

I understand and thanks Joost. I wonder what to listen to at Bert's.

[Joost Buijs]

Krzysztof,

When my engine researches a node (this usually happens in a search) it also reevaluated that node. When the evaluation is very fast this is of no concern, but with the slow NNUE evaluation it is a nuisance.

This morning I fixed this, and now the speed went up from 55 to 88 Mn/s. Very stupid of me that I didn't think about this before, I knew the problem was there but I didn't expect it to have such a big impact. Now that this is fixed it will run somewhat faster on your machine too.

Joost

[Krzysztof Grzelak]

Thank you for the information Joost.

[Ed Gilbert]

Who can from programmers to make the program available for the tournament - please. Many thanks in advance.

Kingsrow version 1.63 available at the download page. It's an incremental change from 1.62, testing about 3 elo better at fast TC.

-- Ed

[Krzysztof Grzelak]

Kingsrow version 1.63 available at the download page. It's an incremental change from 1.62, testing about 3 elo better at fast TC.

-- Ed

Thanks Ed for the info. I changed the versions of the program some time ago.

[Joost Buijs]

Who can from programmers to make the program available for the tournament - please. Many thanks in advance.

Krzysztof,

You can download my engine from Dropbox at the following link:

https://www.dropbox.com/s/dkk7zy3026eo8 ... b.rar?dl=0

Don't expect too much from it, the network remains slow. On my Intel machine using 16 cores it does 60 to 75 Mn/s. I expect it to be slower on the AMD-TR1.

It is compiled for AVX2 with -march=znver1, so I expect it to run without problems on the TR1. It can run about 25% faster with AVX-512, unfortunately that won't work on AMD.

There are 3 things you can edit in the ini-file, threads 1..64, tt_size 26..34 (30 is 1Gb), and the dxp_port (default is 27531).
If you have 'lock pages in memory' enabled on your PC it will automatically use large pages, in this case you can set the tt_size at 32 (4 Gb), otherwise leave it at 30.

With very fast time-controls the engine is clearly weak, this is mainly caused by the slowness of the network. With normal time-controls and multi-core it sits approx. 10 Elo below Kingsrow ver. 1.63, sometimes better, sometimes worse.

Edit: After testing I found out that the version of the network that I put on Dropbox is probably not the best, it scores worse than expected, I will replace it later on with a better version. In the mean time you can test the engine as it is. The problem is that I don't have the resources to test each network with long time-controls because I need my computers for something else.

When I can find the time to add pondering before your tournament starts I will upload a new version.

Let me know if you encounter any problems, in case you do I will try to solve them.

Joost

[Krzysztof Grzelak]

Thank you Joost. I have two questions: I have two questions: how will the " large pages " option affect the program, and how to see the maximum options Mn/s. For me on TR1 maximum options Mn/s this 53.0.

[Joost Buijs]

Thank you Joost. I have two questions: I have two questions: how will the " large pages " option affect the program, and how to see the maximum options Mn/s. For me on TR1 maximum options Mn/s this 53.0.

Large pages help to avoid the slowdown of transposition table access with large tables due to TLB trashing. With tables smaller than 1.0 Gb. this is usually not a problem. Since you are also using chess engines (most of the stronger ones can use large pages), it might be better to turn the "lock pages in memory" option on in Windows.

The 53.0 Mn/s on your TR1 is better than expected, I have no experience with TR1 but I always hear people telling that it has slow AVX2, 53 Mn/s is not bad at all.

I think that the problem with the version I uploaded on Dropbox is not the network itself but the int8 quantization, I switched back to int16 and I'm busy testing it. When it looks fine I will post a new link with the corrected version today.

Joost

[Krzysztof Grzelak]

Thank you Joost. Of course, there is no problem to enable the options " Large Pages ". And I do not interfere with testing.

[Joost Buijs]

The problem with the engine is indeed the 8bit quantization, after switching to 16bit it even has a head start after 48 games. I'll let it run until all the 158 games are finished and post a new link after this. This will take a few hours.

I will remove the "broken" version from Dropbox.