Chess System Tal 2 - Gripes

[chrisw]

Bonjour,

If you are a power user of engines, or of Chess System Tal 2 in particular, please remind me of gripes you may have had about the engine. One gripe was related to limiting threads to number of cores, IIRC. How many threads should it be limited to? Or unlimited? Another gripe was occasional random flake out on specific EPDs. Were there any others?

Thanks in advance.

[peter]

Well, MultiPV would be nice to have, and then there's this old bug in book- usage (since some versions of CST and latest Rebel- engines), that capturing moves aren't played out of book.
Syzygy- tbs- usage could help in analysis now and then too, of course, even if it probably wouldn't give much Elo- gain in engine-engine-game playing, regards

[Frank Quisinsky]

Well, I’ve seen so many CSTal games by now, there are definitely over 15.000 in my database. I find the EAS version particularly interesting, but also the Extreme version. There hasn’t been a single crash.

What’s technically implemented or which features are included is a matter of taste. I don’t think every engine should use the Syzygy endgame databases. Besides, there are also the Gaviota databases and others.

When I see that engines in the top half hardly ever lose games anymore, I lose a bit of interest in testing engines. It’s just boring to watch when I know in advance that the engine won’t lose anyway. Measuring playing strength based on the number of draws, well, I would never have believed that many years ago.

Ultimately, I see enough losses from engines about 100 Elo below Stockfish and Co. To be honest, those are also the engines I prefer to watch. It’s much more fun to watch them.

Whether an engine uses 4, 8, 16, 32, etc., cores is hardly relevant anymore in the age of neural networks, who cares if, at 3400 Elo, an extra core adds 5 Elo, or if 16 more cores might add 30 Elo?

Expectations today are completely overblown, and often I no longer understand why so many programmers are so eager to come as close as possible to Stockfish’s performance.

However, if aggressive engines produce many quick wins, these engines are particularly well-suited for opening analysis. That is the only phase of the game that is still interesting for humans. In the other phases of the game, we have reached near the perfection.

In that respect, the three leading CSTal versions are an achievement because they serve the area that, in my view, is truly interesting specifically, two of the three CSTal versions (CSTal Extreme, CSTal EAS).

What I would like to see is a constant effort to lower the average number of moves, while ensuring that draws that occur too quickly (up to about 25 moves) are avoided and, toward the end of the game, clear draw positions are resolved more quickly. However, CSTal already plays very well.

Again, totay so many things around engines are completly overstimulated. Interesting UCI features is the secret or such things CSTal do, with a very aggressive style of chess.

[Frank Quisinsky]

Forget it, we shouldn’t be telling programmers what they should or shouldn’t do anymore. Each of these gentlemen has enough ingenuity to come up with something beautiful and unique on their own when it comes to exciting UCI features.

Why am I writing this?
Because that’s exactly what makes an engine great: finding something special that other engines don’t have. Basically, these programmers can really let their creativity run wild in this area.

Make the engines great again.

It’s not just about Elo, especially since Elo is so boring, there’s nothing more boring on earth than Elo.

[Frank Quisinsky]

Neural Network It's both a curse and a blessing.

What I like to see are stronger HCE engines.
With the full power of programmers own evaluations with many interesting UCI features.

Example:
Weiss, what for a great engine!
How many work the programmer do in this field.

That's what I mean, in times today a rough diamond!
Or the JAVA engine Calvin ... what for a great style vs. clearly stronger.
To beat Calvin is an order!

Such things ...

[Steve Maughan]

One issue is the crashes in the Shredder GUI. The issue is something to do with positions that start with a "fen" as opposed to "startpos". Here's an example of a crash:

Code: Select all

Wed Jun  3 09:31:47 2026: from Chess System Tal Extreme (2): info depth 19 seldepth 36 time 1457 nodes 1125157 score cp 247 lowerbound pv b6c6 d7b8 hashfull 7 Knps 772242 
Wed Jun  3 09:31:47 2026: from Chess System Tal Extreme (2): info depth 19 seldepth 35 time 1459 nodes 1127736 score cp 244 pv b6c6 c8a6 c6d7 e5e6 d7e8 a6a8 d3d8 a8c6 e8f8 c6d7 d8e8 d7d4 f8g8 d4c3 e8g6 c3b2 g6g5 g1h1 g5h5 h1g1 h5g6 g1h1 g6e6 b2b8 g8h7 b8c7 a5a4 c7c5 e6h3 h1g1 hashfull 7 Knps 772951 
Wed Jun  3 09:31:47 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:31:47 2026: from Chess System Tal Extreme (2): bestmove b6c6 
Wed Jun  3 09:31:48 2026:   to Chess System Tal Extreme (2): ucinewgame
Wed Jun  3 09:31:48 2026:   to Chess System Tal Extreme (2): isready
Wed Jun  3 09:31:48 2026: from Chess System Tal Extreme (2): readyok
Wed Jun  3 09:31:48 2026:   to Chess System Tal Extreme (2): position fen 8/8/8/8/8/8/ppQKPPP1/k7 w - -
Wed Jun  3 09:31:48 2026:   to Chess System Tal Extreme (2): go infinite
Wed Jun  3 09:32:00 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:03 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:06 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:09 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:12 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:15 2026:   to Chess System Tal Extreme (2): quit
Wed Jun  3 09:32:15 2026:   to Chess System Tal Extreme (2): uci
Wed Jun  3 09:32:15 2026: from Chess System Tal Extreme (2): info string Chess System Tal Extreme
Wed Jun  3 09:32:15 2026: from Chess System Tal Extreme (2): info string (C) Chris Whittington 2003-2026
Wed Jun  3 09:32:15 2026: from Chess System Tal Extreme (2): info string Code Base : Chris Whittington
Wed Jun  3 09:32:15 2026: from Chess System Tal Extreme (2): info string NNUE development : Ed Schroder

The multiple "stop" commands are send after the engine crashes.

There are also node-per-second send as "knps", which should be "nps" — this might have been fixed already. Thanks!

— Steve

[jorose]

Considering it is not open source and I can't find a version for Linux, I suppose my primary issue with it is that it won't run on my system at all.
No other complaints so far.

[chrisw]

One issue is the crashes in the Shredder GUI. The issue is something to do with positions that start with a "fen" as opposed to "startpos". Here's an example of a crash:

Code: Select all

Wed Jun  3 09:31:47 2026: from Chess System Tal Extreme (2): info depth 19 seldepth 36 time 1457 nodes 1125157 score cp 247 lowerbound pv b6c6 d7b8 hashfull 7 Knps 772242 
Wed Jun  3 09:31:47 2026: from Chess System Tal Extreme (2): info depth 19 seldepth 35 time 1459 nodes 1127736 score cp 244 pv b6c6 c8a6 c6d7 e5e6 d7e8 a6a8 d3d8 a8c6 e8f8 c6d7 d8e8 d7d4 f8g8 d4c3 e8g6 c3b2 g6g5 g1h1 g5h5 h1g1 h5g6 g1h1 g6e6 b2b8 g8h7 b8c7 a5a4 c7c5 e6h3 h1g1 hashfull 7 Knps 772951 
Wed Jun  3 09:31:47 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:31:47 2026: from Chess System Tal Extreme (2): bestmove b6c6 
Wed Jun  3 09:31:48 2026:   to Chess System Tal Extreme (2): ucinewgame
Wed Jun  3 09:31:48 2026:   to Chess System Tal Extreme (2): isready
Wed Jun  3 09:31:48 2026: from Chess System Tal Extreme (2): readyok
Wed Jun  3 09:31:48 2026:   to Chess System Tal Extreme (2): position fen 8/8/8/8/8/8/ppQKPPP1/k7 w - -
Wed Jun  3 09:31:48 2026:   to Chess System Tal Extreme (2): go infinite
Wed Jun  3 09:32:00 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:03 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:06 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:09 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:12 2026:   to Chess System Tal Extreme (2): stop
Wed Jun  3 09:32:15 2026:   to Chess System Tal Extreme (2): quit
Wed Jun  3 09:32:15 2026:   to Chess System Tal Extreme (2): uci
Wed Jun  3 09:32:15 2026: from Chess System Tal Extreme (2): info string Chess System Tal Extreme
Wed Jun  3 09:32:15 2026: from Chess System Tal Extreme (2): info string (C) Chris Whittington 2003-2026
Wed Jun  3 09:32:15 2026: from Chess System Tal Extreme (2): info string Code Base : Chris Whittington
Wed Jun  3 09:32:15 2026: from Chess System Tal Extreme (2): info string NNUE development : Ed Schroder

The multiple "stop" commands are send after the engine crashes.

There are also node-per-second send as "knps", which should be "nps" — this might have been fixed already. Thanks!

— Steve

Bonjour Steve,
Pretty sure that one is found. Buffer overflow. Rarely actually happened but with phase of moon and following wind and whatever whatever it struck.
Whether or not it ever makes it back into the release is another matter of course. It will need to be confirmed. Thanks.

[Alibaba]

Chess System TAL is a very interesting engine. It can play fantastic games, but unfortunately I can hardly use it for my purposes. This is partly because it doesn't use as many cores as I specify in the settings. What's worse, when I use the engine for engine tournaments, I often experience crashes in both the Shredder GUI and the DF 16 GUI. The crashes occur more frequently with longer time controls, such as 60 minutes per game, and especially when using Permanent Brain.

[chrisw]

Chess System TAL is a very interesting engine. It can play fantastic games, but unfortunately I can hardly use it for my purposes. This is partly because it doesn't use as many cores as I specify in the settings. What's worse, when I use the engine for engine tournaments, I often experience crashes in both the Shredder GUI and the DF 16 GUI. The crashes occur more frequently with longer time controls, such as 60 minutes per game, and especially when using Permanent Brain.

Source I have here is now fixed, certainly for the cores thing, you can now have as any or as few threads as you like (there's an upper thread limit of 64). I did find something that could account for the random crash fails (as they have been described) and fixed it, but with no way to test GUIs that I don't actually have, no way to know for sure. If an upgrade gets a public release, then users will no doubt report if problem persists. No promises about a public upgrade release, might happen, might not.