mmt Vs. Ovyron (G4 D5 BG2)

[mmt]

1. g4 d5 2. Bg2 Bxg4 3. c4 c6 4. Qb3 e6 5. Qxb7 Nd7 6. Nc3 Ne7 7. cxd5 exd5 8. d4 Rb8 9. Qa6 Rb6 10. Qd3 Ng6 11. h3 Be6 12. Nf3 Bd6 13. h4 h5 14. b3 Nf6 15. Bg5 O-O 16. e3 Re8 17. Kf1 Bg4 18. Ne1 Bb4 19. Na4 Rb8 20. Nc2 Be7 21. f3 Be6 22. Nc5 Bc8 23. Kf2 Nd7 24. Ne6 Qa5 25. Bxe7 Rxe7 26. b4 Qb6 27. Ng5 Ba6 28. Qa3 Rbe8 29. Bf1 Bxf1 30. Raxf1 Qc7 31. Qd3 a5 32. a3 axb4 33. axb4 Qd6 34. Rfg1 Nb6 35. Qf5 Nf8 36. Re1 Nc4 37. Nh3 Ra7 38. Qxh5 Ra2 39. Re2 Qe7 40. e4 dxe4 41. fxe4 Qd7 42. Ng5 f6 43. Nf3 Nd6 44. Nfe1 Rxe4 45. Qf3 Qe6 46. Rxe4 Nxe4+ 47. Kg1 Rb2 48. Rh2

[d]5nk1/6p1/2p1qp2/8/1P1Pn2P/5Q2/1rN4R/4N1K1 b - - 3 11
There were several black's moves in this game where there was only one choice that wouldn't nullify the advantage. I doubt an unaided human could've converted this to a win against top programs.

[jp]

There were several black's moves in this game where there was only one choice that wouldn't nullify the advantage.

I hope you've been keeping record of which moves those were, so you can tell us after the game.

5nk1/6p1/2p1qp2/8/1P1Pn2P/5Q2/1rN4R/4N1K1 b - - 3 11

Komodo 13.2, depth 27: -2.18 (...Rb3).

[jp]

Suppose that stockfish search 1.Ra1 Rb8 2.Rc1 Rd8 3.Re1 Rf8 that is a draw by the 50 move rule
earlier to searching 1.Re1 Rf8 that lead to the same position but win because there is no 50 move problem.

1...Rf8 get reduced so it is always lower depth than the depth after 1.Ra1 Rb8 2.Rc1 Rd8 3.Re1 Rf8 so the hash return a draw after 1.Ra1 Rf8 and stockfish can never see the win.

After enough searching time at big depth stockfish turn off pruning by hash tables but in this case the number of lines is huge so even with fast hardware stockfish is going to need many million of years to see the draw by the 50 move rules because searching something like 10^50 lines is impossible.

Note that I do not claim that it is what happens and I did not learn the code of stockfish but it can explain why programs cannot see the shortest mate when they use the 50 move rule.

Yes, I see, though your example really is just losing tempi by shuffling in a way that maybe doesn't apply to the positions we've looked at. Put it this way: taking a detour to get to a key position that you can mate from is one thing; going randomly from one mating position to another, all of them with different, non-intersecting mating lines that look to the clueless human and computer equally unintelligible, is another thing.

If you ignore the 50 move rules things are easier because there are no misleading draw scores by the 50 move rule to hide mates
but I think that there still may be a problem of draw score by repetition that is not correct in the hash table because in another line there is no repetition so I suspect that there may be mates that stockfish never find even if you ignore the 50 move rule.

If the last case never happen in stockfish then it may be interesting to understand why.

For correspondence chess and analysis, there definitely should be options for changing the 50-move rule, just like switching multiPV or to mate-finding mode. And while we're at it, why not 3-fold repetition too? I'd hope these things can be hacked easily. Surely there must be some parameter for the first that is set to 50, which we can just change to 200 if we want.

[Ovyron]

Rh2

1. g4 d5 2. Bg2 Bxg4 3. c4 c6 4. Qb3 e6 5. Qxb7 Nd7 6. Nc3 Ne7 7. cxd5 exd5 8. d4 Rb8 9. Qa6 Rb6 10. Qd3 Ng6 11. h3 Be6 12. Nf3 Bd6 13. h4 h5 14. b3 Nf6 15. Bg5 O-O 16. e3 Re8 17. Kf1 Bg4 18. Ne1 Bb4 19. Na4 Rb8 20. Nc2 Be7 21. f3 Be6 22. Nc5 Bc8 23. Kf2 Nd7 24. Ne6 Qa5 25. Bxe7 Rxe7 26. b4 Qb6 27. Ng5 Ba6 28. Qa3 Rbe8 29. Bf1 Bxf1 30. Raxf1 Qc7 31. Qd3 a5 32. a3 axb4 33. axb4 Qd6 34. Rfg1 Nb6 35. Qf5 Nf8 36. Re1 Nc4 37. Nh3 Ra7 38. Qxh5 Ra2 39. Re2 Qe7 40. e4 dxe4 41. fxe4 Qd7 42. Ng5 f6 43. Nf3 Nd6 44. Nfe1 Rxe4 45. Qf3 Qe6 46. Rxe4 Nxe4+ 47. Kg1 Rb2 48. Rh2 Rb3

[d]5nk1/6p1/2p1qp2/8/1P1Pn2P/1r3Q2/2N4R/4N1K1 w - -

[mmt]

1. g4 d5 2. Bg2 Bxg4 3. c4 c6 4. Qb3 e6 5. Qxb7 Nd7 6. Nc3 Ne7 7. cxd5 exd5 8. d4 Rb8 9. Qa6 Rb6 10. Qd3 Ng6 11. h3 Be6 12. Nf3 Bd6 13. h4 h5 14. b3 Nf6 15. Bg5 O-O 16. e3 Re8 17. Kf1 Bg4 18. Ne1 Bb4 19. Na4 Rb8 20. Nc2 Be7 21. f3 Be6 22. Nc5 Bc8 23. Kf2 Nd7 24. Ne6 Qa5 25. Bxe7 Rxe7 26. b4 Qb6 27. Ng5 Ba6 28. Qa3 Rbe8 29. Bf1 Bxf1 30. Raxf1 Qc7 31. Qd3 a5 32. a3 axb4 33. axb4 Qd6 34. Rfg1 Nb6 35. Qf5 Nf8 36. Re1 Nc4 37. Nh3 Ra7 38. Qxh5 Ra2 39. Re2 Qe7 40. e4 dxe4 41. fxe4 Qd7 42. Ng5 f6 43. Nf3 Nd6 44. Nfe1 Rxe4 45. Qf3 Qe6 46. Rxe4 Nxe4+ 47. Kg1 Rb2 48. Rh2 Rb3 49. Nd3

[d]5nk1/6p1/2p1qp2/8/1P1Pn2P/1r1N1Q2/2N4R/6K1 b - - 5 12

[mmt]

I hope you've been keeping record of which moves those were, so you can tell us after the game.

Not very well, unfortunately.

[Zenmastur]

I hope you've been keeping record of which moves those were, so you can tell us after the game.

Not very well, unfortunately.

This should be easy enough to figure out by reverse analysis

[Zenmastur]

Initial tests with the new drive show that it can sustain more than 3 times what my other drive can. Testing them is tricky. It's hard to find positions that will drive the TB requests to high numbers and keep them their long enough so that you can go to the right screens and record the data. Finding a position that can drive the request high enough for long enough to cause the drive to heat up to the point of throttling seems problematic. But this is a good thing I think. I haven't been able to test its peak yet but it sustains about 16k+ hard page faults a second and over 1.2GB of random reads per second with ease. My other drive could sustain 4k-5k and about 200Mb of random reads. This allow SF to maintain about 80-90 Mnps while heavy requests to 7-man TB files. I would say it's worth the price premium over a "standard" NVMe drive. One other advantage is it has is 8 times the write life. Not sure I need this since TB request are read only operations but I do have apps that beat the hell out of my hard drives with constant read/write cycles. This would be a good drive to run those apps. Now I just need 4 x 4TB drives like this one in a striped array and I'll be good to go!

The NVMe drives are great but they only go up to 2TB and I only have one spot for them on the motherboard. I'm keeping all the 7-piece TBs on regular SSDs (need the NVMe drive for other things) but I'm pretty much out of space on them. That's a good number of nodes/s for SF!

An update on my new drive. I was able to find a few positions that stressed it for extended periods of time. It was able to sustain 35+k page faults a second and 3.5 Gb a second. At this rate it does have a large negative affect on NPS. They dropped to 45M nps. So, for 7-man TB's it looks like you really do need a striped array of fast PCIe M2 cards to get best/better performance from 7-man TBs.

The new 4 TB drives should be out soon. So quad M2s in a striped configuration would give you near enough space for 7-man and "MAYBE" enough speed so hitting 7-man TBs hard won't slow down the CPU so much.

One other note, I've come to the conclusion that 5-man with DTM for general analysis is better than 7-man syzygy and better yet 6-man with DTM would be far superior. I actually have 6-man Nalimov but few programs can use them.

Regards,

Zenmastur

[drewdrew]

Stockfish-dev seems to be doing a bit better here - my apologies for the raw UCI output; if anybody knows of an easy way of converting this to more friendly notation, let me know -:

Thanks, Drew. I wonder why your output is a lot different from Louis's. (The eval movement and fail high, etc.)

Here's Zenmastur's method for notation conversion. I have not tried it yet.

I took a PGN header with a FEN line like this one:

Code: Select all

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[FEN "5B2/1n6/1k6/3K4/B7/8/8/8 w - - 0 1"]

Then I appended the moves "as is" from Zullil's post and saved it as a pgn file. I then loaded it into a chess program. Easy peasy!

Thank you! This indeed works perfectly well.

[mmt]

This should be easy enough to figure out by reverse analysis

It took >15 mins to see how less optimal other moves are for a couple of positions for SF, so we'll need to run it with enough time per move. LC0 sometimes didn't get them on its own at all after hours. That's why I don't see unaided humans converting this to a win.