BUG: Rebel 16.2 & Shredder GUI...

Posts : 6 Join date : 2022-12-28 Location : Florida USA

The new Revel 16.2 engine seems to have problems with the Shredder GUI when analyzing positions that don't start from the starting position. Here's an example:

In this position: "1r4r1/3q1npk/2b1pbnp/Rp1p4/1N1P3P/2PQ1pP1/1K3B2/5B1R w - -" there is a nice mate in 15 for white, starting with Qxg6+.

I turned the debug output on. Here's the ucilog.txt when Rebel 16.2 is given the initial position and then Qxg6 is played — all in analysis mode:

Code:: Sat Feb 18 12:55:23 2023: from Rebel-16.2 (0): info depth 22 seldepth 43 time 5726 nodes 12552681 score cp -424 pv h4h5 f7g5 h5g6 h7h8 f2e3 g8f8 e3f4 b8c8 d3e3 g5e4 f1d3 c6b7 a5a7 h8g8 f4h6 c8c3 h6f4 d7e7 b4c2 f8c8 f4e5 f6e5 d4e5 f3f2 c2d4 e7b4 b2a1
Sat Feb 18 12:55:24 2023: from Rebel-16.2 (0): info depth 23 seldepth 41 time 6336 nodes 13803582 score cp -424 lowerbound pv h4h5 h7h8
Sat Feb 18 12:55:24 2023: from Rebel-16.2 (0): info depth 23 seldepth 36 time 6416 nodes 13957504 score cp -409 lowerbound pv h4h5
Sat Feb 18 12:55:25 2023: to Rebel-16.2 (0): stop
Sat Feb 18 12:55:25 2023: from Rebel-16.2 (0): bestmove h4h5
Sat Feb 18 12:55:25 2023: to Rebel-16.2 (0): position fen 1r4r1/3q1npk/2b1pbnp/Rp1p4/1N1P3P/2PQ1pP1/1K3B2/5B1R w - - moves d3g6
Sat Feb 18 12:55:25 2023: to Rebel-16.2 (0): go infinite
Sat Feb 18 12:55:25 2023: from Rebel-16.2 (0): info depth 1 seldepth 1 time 12 nodes 3 score cp -830 pv h4h5
Sat Feb 18 12:55:25 2023: from Rebel-16.2 (0): info depth 2 seldepth 8 time 12 nodes 143 score cp -394 pv h4h5 f7g5 h5g6 h7h8
Sat Feb 18 12:55:25 2023: from Rebel-16.2 (0): info depth 3 seldepth 8 time 12 nodes 236 score cp -394 pv h4h5 f7g5 h5g6 h7h8
Sat Feb 18 12:55:25 2023: from Rebel-16.2 (0): info depth 4 seldepth 8 time 12 nodes 332 score cp -394 pv h4h5 f7g5 h5g6 h7h8
Sat Feb 18 12:55:25 2023: from Rebel-16.2 (0): info depth 5 seldepth 5 time 12 nodes 624 score cp -409 lowerbound pv h4h5
Sat Feb 18 12:55:25 2023: from Rebel-16.2 (0): info depth 5 seldepth 6 time 12 nodes 687 score cp -358 lowerbound pv h4h5
Sat Feb 18 12:55:25 2023: from Rebel-16.2 (0): info depth 5 seldepth 6 time 12 nodes 738 score cp -350 pv h4h5 f7g5 h5g6 h7h8 a5a6

After the "Stop" command the Rebel responds with "bestmove h4h5" (it didn't find the mate). The GUI then makes the move Qxg6 and gives Rebel the "position" command, followed by "go infinite". As you can see from the output, Rebel starts to calculate but hasn't registered the Qxg6 move and still wants to play h4h5.

I suspect the bug is associated with the FEN string format. FWIW, in Maverick I consume the FEN and then search for the "moves" command i.e., I don't assume the FEN has a fixed number of fields.

Hope it can be fixed easily and I'm happy to test and candidate fixes

— Steve

Posts : 612 Join date : 2020-11-26

It was the case for previous versions too.

Posts : 1254 Join date : 2020-11-17 Location : France

1r4r1/3q1npk/2b1pbnp/Rp1p4/1N1P3P/2PQ1pP1/1K3B2/5B1R w - -

That's a 4-part FEN, I might have been expecting a 6-part, but that ought not to break it. Will take a look.

Posts : 1254 Join date : 2020-11-17 Location : France

Chris Whittington wrote:: 1r4r1/3q1npk/2b1pbnp/Rp1p4/1N1P3P/2PQ1pP1/1K3B2/5B1R w - -

That's a 4-part FEN, I might have been expecting a 6-part, but that ought not to break it. Will take a look.

Yup, it expects a 6-part string and interprets "moves" as an attempt at sending "move number" or "rule50" or whatever it usually is in a 6-part fen. Will fix it tomorrow.

Posts : 1254 Join date : 2020-11-17 Location : France

Done. Code is with Ed for actioning.

Thanks Steve Maughan!

Posts : 6 Join date : 2022-12-28 Location : Florida USA

Great — let me know if you need me to test any fixes.

Sleep well!!

Steve

Subject: Re: BUG: Rebel 16.2 & Shredder GUI... Sun Feb 19, 2023 10:02 am

http://rebel13.nl/windows/rebel-16.html

Silent update.

1. Fixes the Shredder bug Rolling Eyes

since things like these don't happen in Arena.

2. On special request, Evalcorrect maximum is now 500.

Posts : 1254 Join date : 2020-11-17 Location : France

Admin wrote:: http://rebel13.nl/windows/rebel-16.html

Silent update.

1. Fixes the Shredder bug since things like these don't happen in Arena.

2. On special request, Evalcorrect maximum is now 500.

I dread to imagine what that might do. Btw, note for testers, whatever value you give to eval correct, the UCI prints will convert evals back to centipawns, base pawn=100

Posts : 6 Join date : 2022-12-28 Location : Florida USA

I can confirm that this version works flawlessly in the Shredder GUI!

Many thanks,

—Steve

Subject: Re: BUG: Rebel 16.2 & Shredder GUI... Sun Feb 19, 2023 3:41 pm

Ed is there something wrong with Rebel CPU usage?

In my offline analysis there is only 36% of CPU used even though I am using maximum 62 from 64 cores

Another thing I have noticed is that Rebel does not support Threads option. From 62 cores I should

also be able to use 126 Threads but there is no such thing in Rebel. Maximum cores that Rebel can use

is 64 and that is it. Any possibility to fix this ?

Subject: Re: BUG: Rebel 16.2 & Shredder GUI... Sun Feb 19, 2023 3:42 pm

I forgot to mention this happens in Fritz 18 GUI that I am using..

Posts : 1254 Join date : 2020-11-17 Location : France

Damir Desevac wrote:: Ed is there something wrong with Rebel CPU usage?

In my offline analysis there is only 36% of CPU used even though I am using maximum 62 from 64 cores

Another thing I have noticed is that Rebel does not support Threads option. From 62 cores I should

also be able to use 126 Threads but there is no such thing in Rebel. Maximum cores that Rebel can use

is 64 and that is it. Any possibility to fix this ?

No. I have forbidden threads > cores.

Subject: Re: BUG: Rebel 16.2 & Shredder GUI... Sun Feb 19, 2023 7:02 pm

I see. But anyway, from 62 cores to only about 36%-39% of CPU usage... Is this normal ?
Some user on outskirts chess forum wrote that with Rebel 15.1.he is able to achieve maximum CPU usage with it. 97,7% of CPU usage, but not with other more recent Rebel versions.. He gets about 50%

Posts : 6 Join date : 2022-12-28 Location : Florida USA

It's not a bug but Rebel doesn't provide all possible search data. Specifically, Rebel doesn't send: "currmove", "currmovenumber", "nps", and "hashfull". It would be a nice-to-have and relatively easy to add. Here's Stockfish's output:

Code:: Sun Feb 19 09:43:05 2023: from Stockfish 15.1 (1): info depth 30 currmove g1f2 currmovenumber 48
Sun Feb 19 09:43:05 2023: from Stockfish 15.1 (1): info depth 30 currmove d1d7 currmovenumber 49
Sun Feb 19 09:43:05 2023: from Stockfish 15.1 (1): info depth 30 currmove e3a7 currmovenumber 50
Sun Feb 19 09:43:05 2023: from Stockfish 15.1 (1): info depth 30 seldepth 60 multipv 1 score cp 373 nodes 19799371 nps 1516495 hashfull 56 tbhits 0 time 13056 pv g5h7 h8h7 d1h5 h7g8 f4f5 d7d6 f1f3 c6e5 f3h3 e5c4 f5e6 c8e6 e3g5 e7g5 h5g5 e6h3 g5d5 g8h7 d5c4 h3d7 c4e4 d7f5 e4h4 h7g8 h4c4 g8h8 c4c7 f8c8 c7f7 f5c2 a1e1 c8f8 f7d5 c2h7 h2h3 a8e8 e1e8 f8e8 d5b7 h7e4 b7a6 d6d5 a6a7 e8c8

— Steve

Posts : 1254 Join date : 2020-11-17 Location : France

Damir Desevac wrote:: I see. But anyway, from 62 cores to only about 36%-39% of CPU usage... Is this normal ?
Some user on outskirts chess forum wrote that with Rebel 15.1.he is able to achieve maximum CPU usage with it. 97,7% of CPU usage, but not with other more recent Rebel versions.. He gets about 50%

I don't know what normal is.
I do know that if the threads usage is pushed too high, Windows can get unstable, and I don't want to have to deal with complaints from end users all over the forums telling me the engine crashes or hangs or something. Anyway, my tests show that increasing threads over cores, if it does work without crashing, gains nothing.

Posts : 222 Join date : 2021-08-28

I hope that the user is aware that 64 cores means 128 threads, but i don't think so...

Posts : 880 Join date : 2020-11-25 Location : USA

Chris Whittington wrote:

Damir Desevac wrote:: I see. But anyway, from 62 cores to only about 36%-39% of CPU usage... Is this normal ?
Some user on outskirts chess forum wrote that with Rebel 15.1.he is able to achieve maximum CPU usage with it. 97,7% of CPU usage, but not with other more recent Rebel versions.. He gets about 50%

I don't know what normal is.
I do know that if the threads usage is pushed too high, Windows can get unstable, and I don't want to have to deal with complaints from end users all over the forums telling me the engine crashes or hangs or something. Anyway, my tests show that increasing threads over cores, if it does work without crashing, gains nothing.

I found the same results. And in some cases it has a negative effect on play.

Posts : 1254 Join date : 2020-11-17 Location : France

stevemaughan wrote:

It's not a bug but Rebel doesn't provide all possible search data. Specifically, Rebel doesn't send: "currmove", "currmovenumber", "nps", and "hashfull". It would be a nice-to-have and relatively easy to add. Here's Stockfish's output:

Code:: Sun Feb 19 09:43:05 2023: from Stockfish 15.1 (1): info depth 30 currmove g1f2 currmovenumber 48
Sun Feb 19 09:43:05 2023: from Stockfish 15.1 (1): info depth 30 currmove d1d7 currmovenumber 49
Sun Feb 19 09:43:05 2023: from Stockfish 15.1 (1): info depth 30 currmove e3a7 currmovenumber 50
Sun Feb 19 09:43:05 2023: from Stockfish 15.1 (1): info depth 30 seldepth 60 multipv 1 score cp 373 nodes 19799371 nps 1516495 hashfull 56 tbhits 0 time 13056 pv g5h7 h8h7 d1h5 h7g8 f4f5 d7d6 f1f3 c6e5 f3h3 e5c4 f5e6 c8e6 e3g5 e7g5 h5g5 e6h3 g5d5 g8h7 d5c4 h3d7 c4e4 d7f5 e4h4 h7g8 h4c4 g8h8 c4c7 f8c8 c7f7 f5c2 a1e1 c8f8 f7d5 c2h7 h2h3 a8e8 e1e8 f8e8 d5b7 h7e4 b7a6 d6d5 a6a7 e8c8

— Steve

doesn't currmove end up just flooding the UCI?

I think it does send hashfull, it waits a bit, then every ten seconds or so, plus some other stuff

Posts : 6 Join date : 2022-12-28 Location : Florida USA

Chris Whittington wrote:: doesn't currmove end up just flooding the UCI?

I think it does send hashfull, it waits a bit, then every ten seconds or so, plus some other stuff

You would think "currmove" would flood the pipe but in reality it doesn't. I've just done a test: a position with 35 moves, searched to a depth of 30 ply is ~1000 lines to support the "currmove" which slows down the engine by <0.1 seconds. Another option is wait for a short period before starting to broadcast the "currmove" statistics.

— Steve

» SPCC: Testruns of Rebel EAS 2 and Rebel Extreme finished
» Shredder 14
» Deep Shredder 13 - review
» Small Somu 1.7a problem ...
» Broadcasting Shredder Classic 5 vs Wasp 4.50 (1 Thread)