|
< Earlier Kibitzing · PAGE 1 OF 3 ·
Later Kibitzing> |
| Jul-05-20 | | Predrag3141: My thought on this one, after two very rough days with puzzle of the day, was that 17 … f4 sould be hard to handle. It opens a file against the abandoned point, f2, and there is no pleasant way to capture. Once I saw 17 … f4 18 Nxf4, Ngxf2 did not look too appetizing but this did: 17 … f4 18 Nxf4 Rxf4 19 gxf4 (not played) Qh4.
So I went with 18 … Rxf4 19 f3, which I'm still pondering as I write this. |
|
| Jul-05-20 | | Predrag3141: 20 … Nxg2 was the tough move to find. Instead 20 … Nxd1 21 Rxd1 leaves Black up a full rook for the moment, and up the exchange for the long term. But in 6 second analyses, Stockfish rates 20 … Nxg2 about 1.5 pawns better than 20 … Nxd1, which is -1.30. |
|
| Jul-05-20 | | NBZ: The best I could found, unfortunately, was the flawed combination 17. ... Nexf2 with the idea that 18. Nxf2? Ne3 wins back the bishop on g2. However, after the simple 18. Bxb7 Nh3+ 19. Kh1! (not Kf1 Ne3+) the best Black can do is 19. ... Ne3 "winning" a rook for two pieces, which all in all is a great deal for White. |
|
| Jul-05-20 | | Brenin: <Predrag3141>: Yes, 20 ... Nxg2, taking the B rather than the R on d1, was good judgment by Black, as the B was a more valuable part of White's defence than the R at that point. Weakening f2 with 14 Rfd1, and not playing h3 to deter a second N attacking it from g4, was a mistake, and the manoeuvre Nf3-e1-d3 invited the thematic f4 and Rxf4. Instead, eliminating the N on e4 with Nd2 was needed. I know to my cost that White has to play precisely when Black builds up this kind of Dutch K-side pressure. |
|
| Jul-05-20 | | goodevans: After 17...f4 it's pretty obvious that 18.gxf4 Qh5 is terrible for white so with white's pawn rooted to g3 doesn't 18.Nxf4 just give away the N? On the other hand, if white doesn't take on f4 then after ...fxg3; hxg3 black gets huge pressure on the f-file. So if, like me, you saw that much and decided it had to be 17...f4 then I guess you can say you 'sort of ' solved a Sunday puzzle. But hats off to anyone who does the analysis and solves it properly. There are plenty of alternatives for white's 18th move to look at. |
|
| Jul-05-20 | | newzild: Like <NBZ>, I wrongly went for 17...Nexf2 with the idea: a) 18. Nxf2 Bxg2 19. Nxg4 (not 19. Kxg2 Ne3+) Be4 followed by 20...fg, winning a pawn and the initiative. b) 18. Bxb7 Ne3 followed by 19...Nxd1, with two rooks and a pawn vs three minor pieces. |
|
Jul-05-20
 | | OhioChessFan: Anyone who could work all this out is in a different universe than me.... |
|
| Jul-05-20 | | mel gibson: That was too difficult for me.
Stockfish 11 agrees with the text:
17... f4
(17. .. f4 (f5-f4 ♘d3xf4
♖f8xf4 f2-f3 ♘g4-e3 ♕c2-d3 ♘e3xg2 ♔g1xg2 ♕e7-g5 d4-d5 ♘e4-c5 ♕d3-e3 e6xd5
♕e3xe8+ ♖f4-f8 ♗b2-f6 ♕g5xf6 ♕e8-e3 d5xc4 ♖c1xc4 ♕f6-b2 ♕e3-d2 ♕b2xb5
♖c4-c3 ♕b5-d7 ♖c3-a3 ♕d7-f7 ♖a3-e3 ♗b7-c8 ♕d2-b2 ♗c8-e6 ♖d1-d4 g7-g5 ♖e3-a3
h7-h6 ♔g2-g1 ♕f7-f6 ♖d4-d2 ♕f6xb2 ♖d2xb2 ♔g8-g7 ♔g1-f1 ♖f8-h8 h2-h4 ♔g7-f6
♔f1-f2 ♖h8-f8 ♔f2-e1 a5-a4 e2-e4 g5xh4 g3xh4 ♖f8-b8 f3-f4 b6-b5 e4-e5+
d6xe5 f4xe5+ ♔f6xe5 ♖b2-e2+ ♔e5-d6) +5.38/44 560) score for Black +5.38 depth 44 |
|
| Jul-05-20 | | RandomVisitor: Lc0 agrees with <mel gibson>, in this 10 minute run using only my graphics card GPUs. Note the blazing speed of 417 positions evaluated per second. The neural net 'programming' was the same used for the recently concluded TCEC season 18. FEN: 
click for larger viewLc0_0.26.0_384x30-3972-swa-20000.pb:
<13/29 10:15 257k 417 -3.56 17. ... f4 18.Nxf4 Rxf4> 19.f3 Ne3 20.Qd3 Nxg2 21.Kxg2 Qg5 22.Rf1 Ref8 23.Qe3 h5 24.d5 Nc5 25.Kh1 R4f5 26.Qxg5 Rxg5 27.e4 Nd3 28.Rc2 Nxb2 29.Rxb2 13/29 10:15 257k 417 -0.47 17. ... Qg5 18.e3 Qh6 19.h3 Nexf2 20.hxg4 Nxd1 21.Rxd1 Bxg2 22.Kxg2 Qxe3 23.Re1 Qg5 24.Nf4 Qxg4 25.Qe2 Qxe2+ 26.Rxe2 Kf7 27.c5 g5 28.Nd3 h5 29.Rc2 Rc8 13/29 10:15 257k 417 -0.20 17. ... Nexf2 18.Bxb7 Nh3+ 19.Kh1 f4 20.Qd2 Qg5 21.Rf1 Nxh2 22.Rxf4 Nxf4 23.gxf4 Qh5 24.Kg2 Rf6 25.Rh1 Rg6+ 26.Kf2 |
|
| Jul-05-20 | | mel gibson: <Lc0 agrees with <mel gibson>, in this 10 minute run using only my graphics card GPUs. Note the blazing speed of 417 positions evaluated per second. The neural net 'programming' was the same used for the recently concluded TCEC season 18.> That's good -
you can play as Leela Chess Zero. |
|
| Jul-05-20 | | jith1207: Truly insane.
I don't even live in the same universe as <OCF>, so I have no clue how many light years and galaxies and eons away Ulibin was when he worked out all of that. |
|
| Jul-05-20 | | agb2002: Black has a bishop and a knight for the bishop pair. White is about to play d5 since the white queen x-rays the pawn on c7 (f3 is met with Ne3). The position invites to play Qg5 or f4. In the case of 17... f4: A) 18.Nxf4 Rxf4 19.gxf4 Qh4
A.1) 20.Bxe4 Qxf(h)2+ 21.Kh(f)1 Qxh(f)2#.
A.2) 20.h3 Qxf2+ 21.Kh1 Ng3#.
A.3) 20.f3 Qxh2+ 21.Kf1 Ne3+ 22.Ke1 Qg3#.
A.4) 20.e3 Qxh2+ 21.Kf1 Nxg3+
A.4.a) 22.Ke1 Qg1+ 23.Kd2 Qxf2+ 24.Kc3 (24.Kd3 Qxe3#) 24... Ne2+ 25.Kb3 Nxc1+, followed by the capture on g2, wins three pawns. A.4.b) 22.fxg3 Nxe3+ 23.Kf2 (23.Ke1 Nxc2+ wins) 23... Nxc2 24.Rg1 Nd3 25.Kxd3 Qxg3+ 26.Kc2 Bxg2 is winning. B) 19.gxf4 Qh4 20.h3 Nxf2 looks very good for Black. C) 19.d5 fxg3
C.1) 20.Bxe4 gxf2+
C.1.a) 21.Kf1(g2) Ne3+ and 22... Nxc2 wins decisive material. C.1.b) 21.Nxf2 Nxf2 22.Bg2 Nxd1 wins decisive material. C.1.c) 21.Kh1 Qh4 is winning.
C.2) 20.hxg3 Nexf2 wins a pawn at least.
C.3) 20.fxg3 Ne3 21.Qb1 Nxd1 22.Bxe4 (22.Rxd1 exd5) 22... Nxb2 23.Qxb2 exd5 wins an exchange. |
|
| Jul-05-20 | | catlover: This is a Puff Tuzzle. |
|
| Jul-05-20 | | RandomVisitor: One highlight from the recently concluded TCEC season 18 (where Stockfish defeated Lc0) was this queen sacrifice by Lc0 (playing white, vs. Stockfish) in game 65 out of 100 played:
click for larger viewLc0 plays 20.Ne4!? Nxb4 21.Qxb4 Bxb4
Play contined 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Rc7 25.Red1 Re7 26.b5 b6 27.Kh2 Rb7 
click for larger viewBlack is paralyzed.
My version of Lc0 finds 20.Ne4! after 5 minutes of 'thought' on my significantly weaker GPU setup. Here is a 30 minute run: Lc0_0.26.0_384x30-3972-swa-20000.pb:
<18/44 30:00 349k 193 +0.63 20.Ne4 Nxb4 21.Qxb4 Bxb4 22.axb4 Kh8> 23.Nfg5 Qb6 24.Nxf7+ Kg8 25.Nfd6 Rf8 26.Nxc8 Bxc8 27.Bxf8 Kxf8 28.Nf6 Ke7 29.Re4 Qxb4 30.Rexd4 Qc3 31.Rd8 b5 32.f4 g5 33.Rh8 Kf7 34.fxg5 Kg6 35.Rf1 Qd4+ 36.Rf2 Bb7 37.Rg8+ Kf7 38.Rb8 Qa1+ 39.Rf1 Qd4+ 40.Rf2 Qa1+ |
|
| Jul-05-20 | | drollere: like some others, i went for the basic idea 17. .. N(e)xf2 and N(g)d3. |
|
| Jul-05-20 | | unrepentant: YEEEEHA!!! FINALLY! I've had a very tough last coupla weeks with the puzzles. I'm just a mediocre B+ club player, but I had this down to black's 24th move... I dunno why, but it was just one of those things- the odd "vision" made this seemingly easy til I got stuck trying to find the best rook move. My 24th was 24. ... Ref1. Dunno if that would've transposed or what. But this is the first time ever I've "solved" a puzzle into the 6th move. yay! |
|
| Jul-05-20 | | goodevans: I've a bit more time to spare than I had this morning so I thought I'd revisit this puzzle. I was surprised that no one has published any analysis of <18.Bxe4>. It takes a few moves to see that this fails for white. Here's what I would consider to be the main line for this: 17...f4 18.Bxe4 Bxe4 19.Qd2 Bxd3 20.exd3 fxg3 21.fxg3 Rf2 22.Qxf2 Nxf2 23.Kxf2 Rf8+ 24.Kg2 Qg5 ... and black's Q will come to e3 or d2 with devastating effect. |
|
| Jul-05-20 | | Marcelo Bruno: A beautiful blockade game! |
|
Jul-05-20
 | | AylerKupp: <<OhioChessFan> Anyone who could work all this out is in a different universe than me....> Are you trying to establish a new standard for social distancing? |
|
Jul-05-20
| | AylerKupp: <<Random Visitor> Congratulations on getting Lc0 running on your system with GPU support. What kind and how many GPUs did you get to use? And I suppose that your next step is to get an array of Titan RTX GPUs, your analyses will then really be fast indeed. The Titan RTX is said to be ~ 10% faster then the 2080 TI used in the last two TCECs, but if you are able to use tensor cores (think AlphaZero) then it's approximately 2X as fast. See https://www.digitaltrends.com/compu.... Of course, it's also about twice as expensive. But maybe <chessgames.com> can take up a collection. I also enjoyed the "depth" that Lc0 reached in 10 mins, 13 ply. Imagine the depth that Stockfish running on your system could reach in 10 mins of calculation! But I'm sure you're aware that search depth and pawn equivalent evaluations in engines using MCTS are an estimate. I don't know how Lc0's search depths are calculated but in Komodo MCTS this is done on the basis of the number of nodes searched. So, with a "blazing" speed of 417 nodes/sec I would not be surprised that the "classic" search depth estimated was so low. During a recent analysis using my old 32-bit computer using Komodo MCTS with only 1 thread Komodo was able to "achieve" an average of about 45 nodes/sec on a d=25 search. Which means that Lc0 on your system was about 9.3X faster than mine with Komodo MCTS. The evaluation in pawn equivalence is also an estimation and consists of a mapping of (I think) scoring percentage (all the documentation refers to this as "winning percentage" but I think that's incorrect) from the probability range [0.00, 1.00] to a pawn equivalence in a range [-12n, +12n]. The [+\-12n] is an approximate artifact of the way Stockfish reports winning positions based on the information provided by the Syzygy tablebases, where n = [ ±128] [ ± ~ 8]. Other engines report evaluations in the range [ ± 25n] were n = [ ±156] [ ± ~ 8] but { ± 128] is good enough for a conversion range. Be on the lookout for Lc0 reporting evaluations much different than other engines. In its initial implementation the conversion was much more optimistic, typically by a factor of ~ 2X higher, but this was recently corrected to generate more reasonable values. I recently attempted to create a table to correct the "original" MCTS evaluation conversions to the "updated" MCTS evaluation conversions as I described in Nunn vs J A Sutton, 1984 (kibitz #93), but I don't know if I'm correct. If interested, see also http://chessforallages.blogspot.com... . |
|
| Jul-05-20 | | RandomVisitor: <AylerKupp>I am running 768 CUDA cores on my modest GeForce GFX 1050. The temperature is 67 C, but playing with the fan I can drop it to 64 C. 100 percent of the CUDA cores are operating. Only 1 thread on my 8 processing core PC is in use, and I think it is just configuring the CUDA cores and managing the flow of information. As far as the score in centipawns, that is configurable via the Lc0 gui, but few would understand 'winning percentage' so I left it in centipawns. I believe the TCEC machine used for the recent competition had four 2080 graphics cards connected together. A 2080 graphics card will run you about $1200.00. I was on NewEgg and Amazon, just looking... |
|
Jul-05-20
 | | Breunor: I’m with Jith1207 a few universes away. |
|
| Jul-05-20 | | Mudphudder: No matter how many times I look at this puzzle (with the answer) I still can not make sense of it......my mind just can not compute!!!!! |
|
Jul-05-20
 | | An Englishman: Good Evening: Ahhhh, another nice, normal ordinary 5/7 week. Vaguely reassuring—brain not dying yet. Went for 17...Nexf2; 18.Bxb7,Ne3, missing 19.Kxf2,Nxc2; 20.Rxc2 with three pieces for the Queen and perhaps fun for both players. |
|
| Jul-05-20 | | Cheapo by the Dozen: I also went with ... N(e)xf2 on Spielmannesque faith. In this case, the faith turned out to be misplaced. |
|
 |
|
|
< Earlier Kibitzing · PAGE 1 OF 3 ·
Later Kibitzing> |
Kibitzing
icon.