1.g4 opening is losing?

[Zenmastur]

It's funny to see what's going on here.

I only wanted to mention I have 1.g4 down to -1.11 with LC0.
It's a draw.
And I have completely other variations.

Don't post them yet, we're looking for an opponent that would be willing to defend 1.g4, so it's great to have you here. If you've found a drawing line that nobody has we'd love to see it played in a game.

You know, of course, I've planned on playing all the time. Right?

The only question is which line to play.

Do be advised that I have to move my OS to a new drive ( so I may be down for a bit ) AND I still have more hardware to install. If it all fits in the case then I'm good. If it doesn't I'll have to tear it all apart and put it in a new case!

[Zenmastur]

But if he wants to defend 1.g4 I'm game for playing black too.

In fact since this started, you might just as well do one game for all the major lines.

2.g5 and 2.Bg2 will soon be in the books. Which will leave 2.c4, 2.d4, 2.e4, 2.e3, 2.h3 etc. My suspicion is ALL will be white losses.

[Ovyron]

You know, of course, I've planned on playing all the time. Right?

I thought your plan was that in your challenge I posted a mate in 32 and you'd go like "ha! that's 5 moves off, now Ovyron has to defend 1.g4!" I took your messages about being unable to find a single promising endgame as a signal that you'd not want to play the white side!

What surprises me about Hai's claim is that he's talking about low scoring lines, that white can force its way into a line that engines can't push beyond -1.20. I always expected that a draw line would lead to some -2.80 scoring line where black can't make progress, so I suspect Hai has a critical line missing.

[Zenmastur]

You know, of course, I've planned on playing all the time. Right?

I thought your plan was that in your challenge I posted a mate in 32 and you'd go like "ha! that's 5 moves off, now Ovyron has to defend 1.g4!" I took your messages about being unable to find a single promising endgame as a signal that you'd not want to play the white side!

What surprises me about Hai's claim is that he's talking about low scoring lines, that white can force its way into a line that engines can't push beyond -1.20. I always expected that a draw line would lead to some -2.80 scoring line where black can't make progress, so I suspect Hai has a critical line missing.

The problem with Lc0 is it can't see very deep. It's just not fast enough, but who knows, maybe he found a way around that. I've actually been testing Lc0 of late, Just on my CPU as my graphics card is on it's last leg. I do plan on get a new one but I'm waiting for the new RTX 3000 series of cards. Even if they only give the same speed they will draw less power, which is important if you plan on running it 24/7.

On my CPU I'm getting 5k-10k NPS on 128x10 nets, sometimes as high as 20K NPS. On big nets 512x40 it really slows down. I'm lucky to get 100 nps. The small nets are too weak to be useful and the biggest nets are too slow to use. 256x20 nets run at 2K - 5K nps most of the time. So they're next on my test list. The problem is, you can only run 1 computer match at a time (maybe 2 to 4 but thats stretching it). Using AB engines I can run tons of games. I started a 10,000 game SF match with the wrong parameters. It finished in a few minutes. I thought cutechess died. But, it had actually finished the match. I had accidentally set the depth to 3 plies. Which is good if you need a bunch of games for training etc.

Do you know what net mmt is using?

Regards,

Zenmastur

[Ovyron]

The small nets are too weak to be useful and the biggest nets are too slow to use.

Are they? With interactive analysis I've managed to use just 23nps to predict Leela's moves on faster hardware. It's like Leela wants to get some place and you can figure out what place it is, and then a tricycle is fast enough to get there. Specially when Leela wants to lose the game very badly, after showing her the variations she'd have picked a different move in a previous position, presumably a faster net would have seen the losing variation and switched earlier without needing user interaction, but we're still on the same lines.

Still waiting for someone to prove me wrong and beat me by reaching a position where you absolutely need a faster Leela, THEN my hardware upgrade would be the graphics card...

I had accidentally set the depth to 3 plies. Which is good if you need a bunch of games for training etc.

Really? Around here even depth 19 is nothing but garbage... to the point ignoring the data is better than using it in any way...

Do you know what net mmt is using?

No idea, if he told us I forgot.

[mmt]

Just for fun, I've started a game between Stockfish-dev (with Contempt = 0) and Stockfish-dev (with default Contempt = 24). The starting position has Black to move after 1. g4 d5 2. g5. Each side has 20 threads, 32 GB hash and 6-man Syzygy tables. Pondering is disabled. The time control is game-in-12-hours for each side.

Here's where things are now. Probably very few updates before the game is concluded.

1. g4 book d5 book
2. g5 book e5 +1.30/46 2484s
3. d4 -1.21/47 1834s exd4 +1.40/44 217s

Nice, I started one a couple of days ago with a g4 "opening book" I made but with much shorter time controls.

[Zenmastur]

A summary may makes things clearer.

Initial position:

[d]2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36
Three black main moves:

Variation 1:
+M27 36.Rh8+ Kf7 37.Rf1+ Qf5 38.Rxf5+ Kg6 39.Rg5+ Kf6 40.Bf4 axb5 41.Nxa7 Rxa7 42.Rf8+ Rf7 43.Rxc8 Be1 44.Rc6+ Ke7 45.Rc7+ Kf6 46.Rxb5 Rxc7 47.Bxc7 Ke6 48.Re5+ Kd7 49.Rxe1 Kxc7 50.a4 g5 51.Kd2 Kc6 52.Kd3 g4 53.Kxd4 g3 54.Ke5 Kd7 55.Rg1 Kc6 56.Rxg3 Kb6 57.Kd6 Kb7 58.Rg7+ Kc8 59.b3 Kb8 60.Kc6 Ka8 61.Kb6 Kb8 62.Rg8+

Variation 2:
+M26 36.Rh8+ Kf7 37.Rf1+ Qf5 38.Rxf5+ Ke6 39.Re5+ Kd6 40.Bf4 axb5 41.Nxa7 Rxa7 42.Rxc8 Ba5 43.a4 Rf7 44.Rf5+ Kd7 45.Rc7+ Bxc7 46.Rxf7+ Ke6 47.Rxc7 bxa4 48.Rxg7 Kf5 49.Bg3 Ke6 50.Ra7 d3 51.cxd3 a3 52.bxa3 Kf6 53.d4 Kg6 54.d5 Kf5 55.d6 Kg4 56.d7 Kxg3 57.d8Q Kf4 58.Re7 Kf5 59.Qf8+ Kg4 60.Rg7+ Kh5 61.Qh8+

Variation 3:
+M24 36.Rh8+ Kf7 37.Rf1+ Qf6 38.Bc4+ Rd5 39.Bxd5+ Be6 40.Bxe6+ Kxe6 41.Rxf6+ gxf6 42.Nxa7 Be1 43.Bxd4 Kf5 44.Kd1 Bb4 45.Rh6 Ke4 46.Rxf6 Ba5 47.Bb6 Ke5 48.Bxa5 Kxf6 49.c4 Ke6 50.c5 Kf7 51.c6 Kg6 52.c7 Kf5 53.c8Q+ Kf4 54.Kd2 Kg5 55.Qe6 Kh4 56.Qg6 Kh3 57.Ke3 Kh2 58.Kf2 Kh3 59.Qg3+

My claim: the initial position is checkmate in 27 moves at least.
I think I can prove it for any black best defensive moves.

Andreas

Sorry, I didn't get much work done on your lines today. I guess I should post something so here is the first un resolved issue I have with your first solution, the 38. ... Kg6 line:

Code: Select all

[Event "?"]
[Site "?"]
[Date "2020.03.05"]
[Round "?"]
[White "FastGM solution 2"]
[Black "?"]
[Result "*"]
[EventDate "2020.03.05"]
[FEN "2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36"]

36.Rh8+ Kf7
37.Rf1+ Qf5 
38.Rxf5+ Kg6 
39.Rg5+ Kf6 
40.Bf4 axb5 
41.Nxa7 Rxa7 
42.Rf8+ Rf7 

43.Rxc8 Be1 

(  [Stockfish 250220 64 POPCNT] 71:M20 43...Ba5 44.Rxb5 Ke6 45.Re8+ Kd7 46.Re4 Bc7 47.Rb7 Rxf4 48.Rxf4 Kc6 49.Rxc7+ Kxc7 50.a4 Kb6 51.Rxd4 Ka5 52.Rg4 Kb6 53.Rxg7 Kc6 54.Rg5 Kd6 55.Rb5 Kc6 56.c4 Kc7 57.a5 Kd7 58.a6 Kd6 59.a7 Kd7 60.a8=Q Kd6 61.Qa6+ Kd7 62.Rb7+ Kc8 63.Qa8# )
(  [Stockfish 250220 64 POPCNT] 71:M19 43...Be1 44.Rc6+ Ke7 45.Rc7+ Kf6 46.Rxf7+ Kxf7 47.Rxb5 Bf2 48.a4 d3 49.a5 Ke6 50.c4 Kd7 51.Kd2 g5 52.Bxg5 Kc8 53.a6 Ba7 54.Bf4 Bf2 55.Be3 Bh4 56.a7 Kd7 57.a8=Q Bg3 58.Qd5+ Ke7 59.Rb7+ Bc7 60.Rxc7+ Kf6 61.Rf7+ Kg6 62.Qf5# )
(  [Stockfish 250220 64 POPCNT] 74:M19 43...Bb4 44.Rc6+ Ke7 45.Rc7+ Kf6 46.Rxf7+ Kxf7 47.Rxb5 Be1 48.a4 g5 49.Bxg5 d3 50.a5 Ke6 51.cxd3 Kd6 52.Be3 Kc6 53.Rb6+ Kd7 54.a6 Ba5 55.Rb5 Be1 56.a7 Ke7 57.a8=Q Bg3 58.Bg5+ Kf7 59.Qa6 Kg7 60.Rb7+ Kf8 61.Qa8+ Bb8 62.Qxb8# )
(  [Stockfish 250220 64 POPCNT] 71:M17 43...Re7 44.Rc6+ Kf7 45.bxc3 dxc3 46.Be5 Kg8 47.Rc7 Rxc7 48.Bxc7 b4 49.Rb5 Kf7 50.Rxb4 Kf6 51.a4 Ke7 52.a5 Kd7 53.a6 Kxc7 54.a7 Kc6 55.a8=Q+ Kd6 56.Rb5 Ke7 57.Rb6 Kf7 58.Qd8 g6 59.Qd7+ Kg8 60.Rb8# )
(  [Stockfish 250220 64 POPCNT] 71:M17 43...Bxb2+ 44.Kxb2 Re7 45.Rc6+ Kf7 46.Rxb5 d3 47.Rc7 Rxc7 48.Bxc7 d2 49.Rd5 d1=Q 50.Rxd1 g6 51.Rd4 Ke6 52.a4 Kf5 53.a5 Ke6 54.a6 Kf5 55.a7 Kf6 56.a8=Q Kg7 57.Qe4 Kf6 58.Rd6+ Kf7 59.Rd7+ Kg8 60.Qa8# )
(  [Stockfish 250220 64 POPCNT] 71:M17 43...b4 44.Rc6+ Ke7 45.Rc7+ Kf6 46.Rxf7+ Kxf7 47.bxc3 dxc3 48.Rb5 Ke6 49.Rxb4 Kd5 50.a4 g5 51.Bxg5 Kc5 52.Rb8 Kd5 53.a5 Ke5 54.a6 Ke4 55.a7 Kf3 56.a8=Q+ Kg3 57.Be3 Kg4 58.Qe4+ Kg3 59.Rg8+ Kh2 60.Qg2# )

... snip ... snip...

Working from the bottom up,when I got to move 43. I found the move 43. ... Ba5. I didn't work on this too much. But it seems to be a problem. But I haven't finished my analysis of the moves above move 43.

Regards,

Zenmastur

[jp]

I think it also takes a long time to find the shortest mate for 40.Bf4

I think that assisted analysis is basically slower than unassisted analysis.

Do you mean just for that position or generally?

verified all the short mates by chest.

Is Chest still effectively limited to mates in about 13 or shorter?

You know, of course, I've planned on playing all the time. Right?

The only question is which line to play.

Great!

[Zenmastur]

The small nets are too weak to be useful and the biggest nets are too slow to use.

Are they? With interactive analysis I've managed to use just 23nps to predict Leela's moves on faster hardware. It's like Leela wants to get some place and you can figure out what place it is, and then a tricycle is fast enough to get there. Specially when Leela wants to lose the game very badly, after showing her the variations she'd have picked a different move in a previous position, presumably a faster net would have seen the losing variation and switched earlier without needing user interaction, but we're still on the same lines.

They might be OK for predicting a move at low depth, but if your opponent is using a big net and multiple GPU's I don't think it would be of much use. You definitely can't use it to play engine matches with. I tried that and one couldn't mate the other in a KQK endgame!

After watching them do fine in the opening and then blunder away their wins I had seen enough!

Still waiting for someone to prove me wrong and beat me by reaching a position where you absolutely need a faster Leela, THEN my hardware upgrade would be the graphics card...

I guess you need Leela more than I. I can run it on the CPU at speeds fast enough for overnight analysis. e.g. 5000 NPS X 8 hours X 3600 seconds = 144M nodes. So I would actually have to drop the number of cores to something like 6 to keep it from running out of memory. Which means I can run SF on the same position at the same time. So when I wake in thew morning I can have both opinions.

I had accidentally set the depth to 3 plies. Which is good if you need a bunch of games for training etc.

Really? Around here even depth 19 is nothing but garbage... to the point ignoring the data is better than using it in any way...

I guess that depends on how you plan on using the data. If you have enough "garbage" games you can actually extract good data from it. IIRC I only used 5 cores and a few minutes to get 10,000 games. If I used all cores I could generate something like 1,000,000 games over night.

[Zenmastur]

A summary may makes things clearer.

Initial position:

[d]2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36
Three black main moves:

Variation 1:
+M27 36.Rh8+ Kf7 37.Rf1+ Qf5 38.Rxf5+ Kg6 39.Rg5+ Kf6 40.Bf4 axb5 41.Nxa7 Rxa7 42.Rf8+ Rf7 43.Rxc8 Be1 44.Rc6+ Ke7 45.Rc7+ Kf6 46.Rxb5 Rxc7 47.Bxc7 Ke6 48.Re5+ Kd7 49.Rxe1 Kxc7 50.a4 g5 51.Kd2 Kc6 52.Kd3 g4 53.Kxd4 g3 54.Ke5 Kd7 55.Rg1 Kc6 56.Rxg3 Kb6 57.Kd6 Kb7 58.Rg7+ Kc8 59.b3 Kb8 60.Kc6 Ka8 61.Kb6 Kb8 62.Rg8+

Variation 2:
+M26 36.Rh8+ Kf7 37.Rf1+ Qf5 38.Rxf5+ Ke6 39.Re5+ Kd6 40.Bf4 axb5 41.Nxa7 Rxa7 42.Rxc8 Ba5 43.a4 Rf7 44.Rf5+ Kd7 45.Rc7+ Bxc7 46.Rxf7+ Ke6 47.Rxc7 bxa4 48.Rxg7 Kf5 49.Bg3 Ke6 50.Ra7 d3 51.cxd3 a3 52.bxa3 Kf6 53.d4 Kg6 54.d5 Kf5 55.d6 Kg4 56.d7 Kxg3 57.d8Q Kf4 58.Re7 Kf5 59.Qf8+ Kg4 60.Rg7+ Kh5 61.Qh8+

Variation 3:
+M24 36.Rh8+ Kf7 37.Rf1+ Qf6 38.Bc4+ Rd5 39.Bxd5+ Be6 40.Bxe6+ Kxe6 41.Rxf6+ gxf6 42.Nxa7 Be1 43.Bxd4 Kf5 44.Kd1 Bb4 45.Rh6 Ke4 46.Rxf6 Ba5 47.Bb6 Ke5 48.Bxa5 Kxf6 49.c4 Ke6 50.c5 Kf7 51.c6 Kg6 52.c7 Kf5 53.c8Q+ Kf4 54.Kd2 Kg5 55.Qe6 Kh4 56.Qg6 Kh3 57.Ke3 Kh2 58.Kf2 Kh3 59.Qg3+

My claim: the initial position is checkmate in 27 moves at least.
I think I can prove it for any black best defensive moves.

Andreas

Sorry, I didn't get much work done on your lines today. I guess I should post something so here is the first un resolved issue I have with your first solution, the 38. ... Kg6 line:

Code: Select all

[Event "?"]
[Site "?"]
[Date "2020.03.05"]
[Round "?"]
[White "FastGM solution 2"]
[Black "?"]
[Result "*"]
[EventDate "2020.03.05"]
[FEN "2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36"]

36.Rh8+ Kf7
37.Rf1+ Qf5 
38.Rxf5+ Kg6 
39.Rg5+ Kf6 
40.Bf4 axb5 
41.Nxa7 Rxa7 
42.Rf8+ Rf7 

43.Rxc8 Be1 

(  [Stockfish 250220 64 POPCNT] 71:M20 43...Ba5 44.Rxb5 Ke6 45.Re8+ Kd7 46.Re4 Bc7 47.Rb7 Rxf4 48.Rxf4 Kc6 49.Rxc7+ Kxc7 50.a4 Kb6 51.Rxd4 Ka5 52.Rg4 Kb6 53.Rxg7 Kc6 54.Rg5 Kd6 55.Rb5 Kc6 56.c4 Kc7 57.a5 Kd7 58.a6 Kd6 59.a7 Kd7 60.a8=Q Kd6 61.Qa6+ Kd7 62.Rb7+ Kc8 63.Qa8# )
(  [Stockfish 250220 64 POPCNT] 71:M19 43...Be1 44.Rc6+ Ke7 45.Rc7+ Kf6 46.Rxf7+ Kxf7 47.Rxb5 Bf2 48.a4 d3 49.a5 Ke6 50.c4 Kd7 51.Kd2 g5 52.Bxg5 Kc8 53.a6 Ba7 54.Bf4 Bf2 55.Be3 Bh4 56.a7 Kd7 57.a8=Q Bg3 58.Qd5+ Ke7 59.Rb7+ Bc7 60.Rxc7+ Kf6 61.Rf7+ Kg6 62.Qf5# )
(  [Stockfish 250220 64 POPCNT] 74:M19 43...Bb4 44.Rc6+ Ke7 45.Rc7+ Kf6 46.Rxf7+ Kxf7 47.Rxb5 Be1 48.a4 g5 49.Bxg5 d3 50.a5 Ke6 51.cxd3 Kd6 52.Be3 Kc6 53.Rb6+ Kd7 54.a6 Ba5 55.Rb5 Be1 56.a7 Ke7 57.a8=Q Bg3 58.Bg5+ Kf7 59.Qa6 Kg7 60.Rb7+ Kf8 61.Qa8+ Bb8 62.Qxb8# )
(  [Stockfish 250220 64 POPCNT] 71:M17 43...Re7 44.Rc6+ Kf7 45.bxc3 dxc3 46.Be5 Kg8 47.Rc7 Rxc7 48.Bxc7 b4 49.Rb5 Kf7 50.Rxb4 Kf6 51.a4 Ke7 52.a5 Kd7 53.a6 Kxc7 54.a7 Kc6 55.a8=Q+ Kd6 56.Rb5 Ke7 57.Rb6 Kf7 58.Qd8 g6 59.Qd7+ Kg8 60.Rb8# )
(  [Stockfish 250220 64 POPCNT] 71:M17 43...Bxb2+ 44.Kxb2 Re7 45.Rc6+ Kf7 46.Rxb5 d3 47.Rc7 Rxc7 48.Bxc7 d2 49.Rd5 d1=Q 50.Rxd1 g6 51.Rd4 Ke6 52.a4 Kf5 53.a5 Ke6 54.a6 Kf5 55.a7 Kf6 56.a8=Q Kg7 57.Qe4 Kf6 58.Rd6+ Kf7 59.Rd7+ Kg8 60.Qa8# )
(  [Stockfish 250220 64 POPCNT] 71:M17 43...b4 44.Rc6+ Ke7 45.Rc7+ Kf6 46.Rxf7+ Kxf7 47.bxc3 dxc3 48.Rb5 Ke6 49.Rxb4 Kd5 50.a4 g5 51.Bxg5 Kc5 52.Rb8 Kd5 53.a5 Ke5 54.a6 Ke4 55.a7 Kf3 56.a8=Q+ Kg3 57.Be3 Kg4 58.Qe4+ Kg3 59.Rg8+ Kh2 60.Qg2# )

... snip ... snip...

Working from the bottom up,when I got to move 43. I found the move 43. ... Ba5. I didn't work on this too much. But it seems to be a problem. But I haven't finished my analysis of the moves above move 43.

Regards,

Zenmastur

I started a search of this position with a large TT and it resolved the issue I posted above in about 30 minutes...

So, I'll continue vetting the upper portion.