![[it]](/mwf/v12/flags/it.png "Italy")
I read that Rybka has this mysterious (for me) multi-PV bug
What is this?
Any practical example?
Thanks in advance
What is this?
Any practical example?
Thanks in advance
![[de]](/mwf/v12/flags/de.png "Germany")
1. Double click your Rybka engine (e.g. Rybka 4 w32.exe)
2. type "uci" (engine responds with a bunch of options, followed by "uciok")
3. type "isready" (engine responds with "readyok")
3. type "setoption name MultiPV value 3" (engine responds with "info string Displaying 3 variations.")
4. type "position fen
5. type "go depth 17"
6. wait for the engine to stall. It will probably happen within a few seconds.
2. type "uci" (engine responds with a bunch of options, followed by "uciok")
3. type "isready" (engine responds with "readyok")
3. type "setoption name MultiPV value 3" (engine responds with "info string Displaying 3 variations.")
4. type "position fen
1qrb2k1/3b1p1p/B2p1np1/n2Pp3/1p2P3/1P2BNNP/R4PP1/3Q2K1 b - - 0 1
"5. type "go depth 17"
6. wait for the engine to stall. It will probably happen within a few seconds.
If I paste that position in ChessBase Fritz 11 GUI and use Deep Rybka 4 how can I see that Multi Position Variation (MultiPV?) bug in action?
![[is]](/mwf/v12/flags/is.png "Iceland")
> 6. wait for the engine to stall. It will probably happen within a few seconds.
I gave up waiting after around 90 minutes and didn't encounter this problem. Rybka had reached depth 21. I used Aquarium instead of manually typing in the commands.
[+0.53] d=21 1...Rc3 2.Bd2 Rc5 3.Bd3 Rc8 4.Bc2 Qb5 5.Qe1 Rb8 6.Qb1 Qb6 7.Be3 Qb7 8.Ne1 h5 9.Nf3 Be8 (1:33:25) 332892kN [+0.73] d=20 1...Nb7 2.Qe1 Rc3 3.Bc4 Kg7 4.Qa1 Qc7 5.Ra8 Rxe3 6.fxe3 Na5 7.Qe1 Nxc4 8.bxc4 Qb6 9.Rxd8 Qxd8 10.Qxb4 Qc7 11.c5 dxc5 12.Qc3 h5 13.Qxe5 Qxe5 14.Nxe5 Bb5 15.h4 c4 16.Kf1 c3 (1:17:20) 292894kN [+2.06] d=20 1...h5 2.Bxc8 Qxc8 3.Qd3 Qb7 4.Bd2 Bb5 5.Qc2 Bd7 6.Nf1 Qb5 7.Bg5 Qb6 8.Ne3 Bb5 9.Nd2 (1:19:45) 299019kN
So neither of you followed the instructions, but can't reproduce it? Strange... 
I just tried it again from the .exe itself and it stalled within about 5 minutes. Maybe it has something to do with memory layout, or maybe, just maybe, with the fact that you weren't following instructions? Vas has confirmed this bug, though, so it's not just me.
Anyhow, R4 stalls out for me in Infinite Analysis under Aquarium 4 with MPV enabled (3) sooner or later (strangely enough, even with MPV set to 3, I only see two variations. Is my delta too low? Ah yes, that was it...).
I can't make it stall under CB 11 (yet), but I think that you can probably just try it a few times and you'll get hit by it eventually.
I'm on a 1CPU machine, but I think you can enjoy this bug on any configuration.
I just tried it again from the .exe itself and it stalled within about 5 minutes. Maybe it has something to do with memory layout, or maybe, just maybe, with the fact that you weren't following instructions? Vas has confirmed this bug, though, so it's not just me.Anyhow, R4 stalls out for me in Infinite Analysis under Aquarium 4 with MPV enabled (3) sooner or later (strangely enough, even with MPV set to 3, I only see two variations. Is my delta too low? Ah yes, that was it...).
I can't make it stall under CB 11 (yet), but I think that you can probably just try it a few times and you'll get hit by it eventually.
I'm on a 1CPU machine, but I think you can enjoy this bug on any configuration.
So this bug only shows up under very rare conditions.
Shows up all the time and in CB11.
The problem is: this bug shows up frequently if you've got only few cores. With more cores it gets increasingly difficult to reproduce it. Your position may work for single core, but with 8 cores I didn't get a single stall in several tries.
Rybka 4 hangs in multi-Var Modus.
which means more than one Variant to display in Engine Window.
It hangs after some time, and hangs... and hangs...
which means more than one Variant to display in Engine Window.
It hangs after some time, and hangs... and hangs...
![[ge]](/mwf/v12/flags/ge.png "Georgia")
best move b2d4 ponder e7e6
info depth 17 time 160 sec nodes 11177977 nps 69662
engine didnot stall
info depth 17 time 160 sec nodes 11177977 nps 69662
engine didnot stall
![[mx]](/mwf/v12/flags/mx.png "Mexico")
> Any practical example?
No, facing it requires some interaction.
Try following the instructions here:
http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=253489#
Apparently Rybka may "die" and stall forever, though she's not really dead as she'll respond to commands.
Thanks
![[us]](/mwf/v12/flags/us.png "United States")
Yep. Stall.
For me it usually happens after I increase the number of lines after it has already been running by pressing the "+". It only happens maybe one out of 30 times. The nice thing is that if it gets beyond a certain depth it become unlikely so you can walk away and not worry. If it does not happen in the fist several minutes when untouched, It is not likely it will. It is not obvious when it is happening, unless you are familiar with it. It stalls but appears to still be running. You know it is stuck because it can sit for hours on say depth 14. The kn/s keeps changing and the Task Manager still shows it at 100% but it is never going to go to the next ply. You can restart it and it quickly goes past the depth 14 so it is not just hitting a search horizon or something.
It is far less likely to happen if you go the the number of lines you want, then reload the engine changing the hash by adding one K or something.
R4 can still crash for other reasons but it is fairly reliable. The other crashes I have experienced are that one thread will crash then it plays along fine...even giving the the same kn/s rate rather than 3/4 of it...but a full crash is emanate which means that it looses on time.
It also has a bug that with "TC Max Move Time" over 100, it will frequently take too much time the last move of a time control, if it determines that, that position warrants the condition of more time oblivious of the fact that it is overstepping what is allotted (or at least that is my hypotheses as to what is going on...it has happened over 300 times to me in my tournaments often in happens in closed positions it thinks it is winning but are actually likely drawn but can happen in a wide variety of positions).
But the biggest frustration is its inability to give reasonable evaluations in endgames or approaching them. This is vary dangerous for postal players in leagues that permit engine use or really anyone who wants a real evaluation before playing a position to its end. Even with 40 GB of 6-man and less-man tables on flash it is still giving very inaccurate evals. With a Terabyte it may be better but flash is expensive and load times will be nasty. Without flash engine slowdown will be enormous. With flash "Frequently" is not frequently enough. Other engines like Shredder on default settings make many more table queries than R4 on "frequent" and do better (when the tables are on flash) in the endgame. I am almost certain R4 could do much better if it looked up more positions when the tables are on flash. It needs a wider range of choice: One for RAM-disk, one for flash (SSD/flash card/USB thumb drive), one for fast conventional disk/RAID, one for run-of-the-mill hard disk, and one for CD/DVD/Blue ray.
Of course adding some knowledge or having a version that has the knowledge would be welcome too.
It is far less likely to happen if you go the the number of lines you want, then reload the engine changing the hash by adding one K or something.
R4 can still crash for other reasons but it is fairly reliable. The other crashes I have experienced are that one thread will crash then it plays along fine...even giving the the same kn/s rate rather than 3/4 of it...but a full crash is emanate which means that it looses on time.
It also has a bug that with "TC Max Move Time" over 100, it will frequently take too much time the last move of a time control, if it determines that, that position warrants the condition of more time oblivious of the fact that it is overstepping what is allotted (or at least that is my hypotheses as to what is going on...it has happened over 300 times to me in my tournaments often in happens in closed positions it thinks it is winning but are actually likely drawn but can happen in a wide variety of positions).
But the biggest frustration is its inability to give reasonable evaluations in endgames or approaching them. This is vary dangerous for postal players in leagues that permit engine use or really anyone who wants a real evaluation before playing a position to its end. Even with 40 GB of 6-man and less-man tables on flash it is still giving very inaccurate evals. With a Terabyte it may be better but flash is expensive and load times will be nasty. Without flash engine slowdown will be enormous. With flash "Frequently" is not frequently enough. Other engines like Shredder on default settings make many more table queries than R4 on "frequent" and do better (when the tables are on flash) in the endgame. I am almost certain R4 could do much better if it looked up more positions when the tables are on flash. It needs a wider range of choice: One for RAM-disk, one for flash (SSD/flash card/USB thumb drive), one for fast conventional disk/RAID, one for run-of-the-mill hard disk, and one for CD/DVD/Blue ray.
Of course adding some knowledge or having a version that has the knowledge would be welcome too.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill