LCZero Accomplishments and Goals Thus Far

[hgm]

What computer-chess afficidonados will buy is driven by what engines require. They used to by highly-overclocked dual-socket (or more) systems with the best CPUs they could find, and don't care about the graphics. When NN-based engines will be the dominant technology they will by PCs with modest CPUs and the most expensive GPU card. This could actually be cheaper. In the end the only thing that matters is how much Elo you buy for your buck; there is no law that says Chess must be restricted to CPUs.

Of course this will be a bit hard to swallow for those having invested k$20+ on 32 cores and no graphics card, while a basic $1000 PC with a k$5 GPU would play better Chess.

But any 'hardware avantage' should be measured in $$$, (possibly including the price of electricity);, other ways just don't make any sense.

[duncan]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists.

If it surpasses stockfish and becomes the best computer chess entity, (even though 'unfairly') can I wake you, or would it be one big bore and you would prefer to sleep through it.?

As far as I am concerned, it is still a 2100 elo engine there.
I see so many excited people giving a hardware advantage to LCzero, like CCLS does for instance uses a GPU for LCzero and single core CPU for the rest of the engines.

I don't think anybody misses the fact that given a hardware advantage that will basically make the evaluation free, you can increase your elo to your satisfaction.

If Stockfish had its evaluation FGPA'ed it may be a 4000 elo engine but who cares for that anyway ?
Daniel

(1)what is the motivation of chess programmers. Is it an intellectual goal to use their skill to see how much elo they can get with limited resources.
or is it a love of chess that motivates them to see the best chess entity they can get ? The code being just a means to an end

(2)for balance list please could you list the positives of lCzero

[duncan]

1. To reach the level of A0 (estimated 3300 rating in GTX 1060)

did ao uses eqivalent of 16,000 years. ?

[Daniel Shawul]

What computer-chess afficidonados will buy is driven by what engines require. They used to by highly-overclocked dual-socket (or more) systems with the best CPUs they could find, and don't care about the graphics. When NN-based engines will be the dominant technology they will by PCs with modest CPUs and the most expensive GPU card. This could actually be cheaper. In the end the only thing that matters is how much Elo you buy for your buck; there is no law that says Chess must be restricted to CPUs.

Of course this will be a bit hard to swallow for those having invested k$20+ on 32 cores and no graphics card, while a basic $1000 PC with a k$5 GPU would play better Chess.

But any 'hardware avantage' should be measured in $$$, (possibly including the price of electricity);, other ways just don't make any sense.

The AlphaZero paper claims a miracle one algorithm for all (go,chess,shogi) games, even with some presentation of their's putting a huge X on all the heuristic algorithms that Stockfish used. With such emphatic claims, one would be interested in the algorithmic breakthroughs. I actually thought they made a real breakthrough better than Deeblue's. On a closer inspection though, I expected that should have some solutions, for instance, for the tactical problems of MCTS, which they have given references to btw! Apparently, there is no algorithmic breakthrough, all they did is use a huge hardware to hide the problems. That is disappointing to say the least.

I like Giraffe's work better in this regard. He trained a neural network of the appropriate size (that is not too big or too small), showed that the result from NN could actually be better than hand-written eval, measured its stength and reported a 2600 elo on a single core. That is more honest than beating stockfish using a crappy algorithm run on a massive hardware.

[hgm]

Again you refer to 'massive hardware'. But the hardware of AlphaZero was not really massive. Just different. AlphaZero needs fewer transistors to run at 3500 Elo than Stockfish needs to run at 3400 Elo, at 'standard' TC. It is just that the transistors have to be connected in an entirely different way.

That the AlphaZero hardware can do some things 100 times faster than equivalent (in terms of number of transistors) x64 CPUs, doesn't mean a thing. My rowing boat is infinitely faster than my Ferrari, for crossing the English Channel.

[Nay Lin Tun]

I am really happy with Leela's achievements my current hardware. I play RPG games, so I have i5 3ghz 4 cores with 1060 GTX.

Here is, 4 cores stockfish latest version drew vs leela ID 227 ,draw. (overall score in 4 games= 3.5 -0.5 in favour of stockfish, so elo difference is around 300)


https://lichess.org/k5ZPaRqg

[pgn]1. e4 {(e2-e4 c7-c5 c2-c3 Ng8-f6 e4-e5 Nf6-d5 d2-d4 c5xd4 c3xd4 d7-d6 e5xd6
e7-e6 Nb1-c3 Bf8xd6 Ng1-f3 O-O Bf1-d3 Nb8-c6 O-O h7-h6) +0.04/18 2} e6
{(e7-e6 d2-d4 d7-d5 Nb1-d2 c7-c5 e4xd5 e6xd5 Bf1-b5+ Bc8-d7 Qd1-e2+ Qd8-e7
Bb5xd7+ Nb8xd7 d4xc5 Qe7xe2+ Ng1xe2 Bf8xc5 Nd2-b3 Bc5-b6 a2-a4 Ng8-f6
Bc1-f4 Ra8-c8 a4-a5 Bb6-c5 O-O O-O Ne2-d4 Rf8-e8 Rf1-e1 Re8-e4 Re1xe4
d5xe4) -0.17/23 5} 2. d4 {(d2-d4 d7-d5 Nb1-c3 Bf8-b4 e4xd5 e6xd5 a2-a3
Bb4xc3+ b2xc3 Ng8-f6 Bf1-d3 O-O Ng1-e2 Rf8-e8 O-O h7-h6 Ne2-g3 c7-c5 d4xc5)
+0.08/19 2} d5 {(d7-d5 Nb1-d2 c7-c5 Ng1-f3 Ng8-f6 e4xd5 e6xd5 Bf1-b5+
Bc8-d7 Bb5xd7+ Nb8xd7 O-O Bf8-e7 d4xc5 Nd7xc5 Nd2-b3 Nc5-e4 Nf3-d4 O-O
c2-c3 Rf8-e8 Nd4-f5 Be7-f8 Nb3-d4 Bf8-c5) -0.06/24 2} 3. Nc3 {(Nb1-c3
Bf8-b4 e4xd5 e6xd5 a2-a3 Bb4xc3+ b2xc3 Ng8-f6 Bf1-d3 O-O Ng1-e2 Rf8-e8 O-O
h7-h6 Ne2-g3 c7-c5 d4xc5 Nb8-d7 Bc1-e3) +0.09/19 2} Nf6 {(Ng8-f6 e4xd5
e6xd5 Ng1-f3 Bf8-d6 Bf1-d3 O-O O-O a7-a6 Nc3-e2 Rf8-e8 c2-c3 Bc8-g4 Ne2-g3
c7-c5 d4xc5 Bd6xc5 b2-b4 Bc5-b6 h2-h3 Qd8-c7 h3xg4) +0.05/20 1} 4. Bg5
{(Bc1-g5 d5xe4 Nc3xe4 Bf8-e7 Ne4xf6+ Be7xf6 Bg5xf6 Qd8xf6 Ng1-f3 c7-c5
d4xc5 Qf6xb2 Bf1-d3 Qb2-c3+ Nf3-d2 Nb8-d7 O-O Nd7xc5 Bd3-b5+ Bc8-d7)
+0.02/18 1} Bb4 {(Bf8-b4 e4-e5 h7-h6 e5xf6 h6xg5 f6xg7 Rh8-g8 h2-h4 g5xh4
Qd1-g4 Qd8-f6 Qg4xh4 Qf6xg7 Ng1-f3 Nb8-d7 Rh1-h3 a7-a6 Rh3-g3 Qg7-h8 O-O-O
Bb4-e7 Rg3xg8+ Qh8xg8 Qh4-f4 c7-c5 g2-g3 b7-b5 Kc1-b1 b5-b4 Nc3-e2)
-0.09/24 11} 5. e5 {(e4-e5 h7-h6 Bg5-h4 g7-g5 Bh4-g3 Nf6-e4 Ng1-e2 c7-c5
a2-a3 Bb4-a5 d4xc5 Nb8-c6 b2-b4 Nc6xb4 a3xb4 Ba5xb4 Qd1-d4 Bb4xc5 Qd4-a4+
Bc8-d7 Qa4-b3 Ne4xg3 h2xg3) +0.20/18 2} h6 {(h7-h6 Bg5-h4 g7-g5 Bh4-g3
Nf6-e4 Ng1-e2 h6-h5 f2-f3 Ne4xc3 Ne2xc3 c7-c5 a2-a3 c5xd4 Qd1xd4 Nb8-c6
Bf1-b5 Bb4-e7 h2-h4 Bc8-d7 Qd4-d2 g5xh4 Bg3-f4 f7-f6 O-O-O Nc6xe5 Rh1xh4
Be7-d6) -0.10/24 2} 6. exf6 {(e5xf6 h6xg5 f6xg7 Rh8-g8 h2-h4 g5xh4 Qd1-g4
Qd8-f6 Qg4xh4 Qf6xg7 O-O-O Bb4-e7 Qh4-h2 c7-c6 Ng1-f3 Nb8-d7 Bf1-d3 Qg7xg2)
+0.20/20 3} hxg5 {(h6xg5 f6xg7 Rh8-g8 a2-a3 Bb4xc3+ b2xc3 Qd8-f6 h2-h4
g5xh4 Qd1-h5 Nb8-c6 Ng1-f3 Qf6xg7 Qh5xh4 f7-f6 Qh4-h5+ Qg7-f7 Ra1-b1 b7-b6
Bf1-b5 Bc8-d7 g2-g4 Nc6-d8 Bb5-d3 e6-e5 Bd3-f5 e5-e4 Nf3-d2 Bd7xf5 Qh5xf5
Qf7-e6 Ke1-e2 Qe6xf5 g4xf5 Nd8-c6 Rh1-h6 Ke8-f7 Rh6-h7+ Rg8-g7) -0.17/25 2}
7. fxg7 {(f6xg7 Rh8-g8 h2-h4 g5xh4 Qd1-g4 Qd8-f6 Qg4xh4 Qf6xg7 O-O-O Bb4-e7
Qh4-h2 c7-c6 Ng1-f3 Nb8-d7 Bf1-d3 Qg7xg2 Qh2-f4) +0.24/19 1} Rg8 {(Rh8-g8
a2-a3 Bb4xc3+ b2xc3 Qd8-f6 h2-h4 g5xh4 Qd1-h5 Nb8-c6 Ng1-f3 Qf6xg7 Qh5xh4
f7-f6 Qh4-h5+ Qg7-f7 Ra1-b1 b7-b6 Ke1-d2 Bc8-d7 g2-g3 Nc6-a5 Rb1-e1 Qf7xh5
Rh1xh5 Ke8-f7 Bf1-d3 Na5-c4+ Bd3xc4 d5xc4 Rh5-h7+ Rg8-g7) -0.17/25 1} 8. h4
{(h2-h4 g5xh4 Qd1-g4 Qd8-f6 Qg4xh4 Qf6xg7 O-O-O Bb4-e7 Qh4-h2 c7-c6 Ng1-f3
Nb8-d7 Bf1-d3 Qg7xg2 Qh2-f4 Qg2-g4) +0.24/21 2} gxh4 {(g5xh4 Qd1-g4 Qd8-f6
Rh1xh4 Nb8-c6 O-O-O Rg8xg7 Rh4-h8+ Bb4-f8 Qg4-f3 Qf6-g5+ Kc1-b1 Bc8-d7
Ng1-h3 Qg5-g4 Qf3-e3 O-O-O Nh3-f4 Qg4-g5 g2-g3 Kc8-b8 Bf1-e2 Rg7-g8 Rh8-h7
Rg8-g7 Rh7-h8) 0.00/22 1} 9. Qg4 {(Qd1-g4 Qd8-f6 Qg4xh4 Qf6xg7 O-O-O Bb4-e7
Qh4-h2 c7-c6 Ng1-f3 Nb8-d7 Bf1-d3 Qg7xg2 Qh2-f4 Qg2-g4 Qf4-e3 Be7-g5
Nf3xg5) +0.16/21 3} Qf6 {(Qd8-f6 O-O-O Bb4xc3 b2xc3 Rg8xg7 Qg4xh4 Qf6xh4
Rh1xh4 b7-b6 g2-g3 Bc8-b7 c3-c4 Nb8-d7 c4xd5 Bb7xd5 Ng1-e2 Nd7-f6 Ne2-c3
Bd5-f3 Bf1-e2 Bf3xe2 Nc3xe2 Ke8-e7 c2-c4 Rg7-g8 Kc1-c2 Rg8-h8 a2-a4 Rh8xh4)
+0.01/24 1} 10. Qxh4 {(Qg4xh4 Qf6xg7 O-O-O Bb4-e7 Qh4-h2 c7-c6 Ng1-f3
Nb8-d7 Bf1-d3 Qg7xg2 Qh2-f4 Qg2-g4 Qf4-e3 Be7-g5 Nf3xg5 Qg4xg5 f2-f4 Qg5-g3
Qe3-d2) +0.13/20 2} Qxg7 {(Qf6xg7 a2-a3 Bb4-e7 Qh4-f4 a7-a6 Ng1-f3 Nb8-c6
O-O-O Be7-d6 Qf4-e3 Nc6-e7 Bf1-d3 Rg8-h8 g2-g3 Bc8-d7 Qe3-g5 Qg7xg5+ Nf3xg5
Rh8-g8 Ng5-h7 O-O-O Nh7-f6 Rg8-h8 Bd3-e2 Rh8xh1 Rd1xh1 c7-c5 Nf6xd7 Kc8xd7
d4xc5 Bd6xc5) +0.07/24 1} 11. Nf3 {(Ng1-f3 Nb8-c6 O-O-O Bc8-d7 Nc3-b5
Ra8-c8 Qh4-f4 a7-a6 Nb5xc7+ Ke8-d8 Nc7xa6 b7xa6 Bf1xa6 Rc8-a8 Ba6-b7
Ra8xa2) +0.17/20 3} Bd7 {(Bc8-d7 a2-a3 Bb4-e7 Qh4-f4 Be7-d6 Qf4-e3 Bd7-c6
O-O-O a7-a6 Nc3-a2 Bc6-a4 Na2-c3 Ba4-c6) 0.00/26 4} 12. O-O-O {(O-O-O
Nb8-c6 Nc3-b5 Ra8-c8 a2-a3 Bb4-e7 Qh4-f4 a7-a6 Nb5xc7+ Ke8-d8 Nc7xa6 b7xa6
Bf1xa6 Rc8-b8 Ba6-d3 Rb8-b6 Rh1-h7 Qg7-f6) +0.12/19 2} Nc6 {(Nb8-c6 a2-a3
Bb4-e7 Qh4-f4 O-O-O Nc3-b5 Be7-g5 Qf4xg5 Qg7xg5+ Nf3xg5 Rg8xg5 g2-g3 Rg5-g8
Rh1-h4 Rg8-h8 Bf1-e2 Rh8xh4 g3xh4 Rd8-h8 h4-h5 f7-f6 Nb5-c3 Bd7-e8 b2-b4
Nc6-e7 Rd1-h1 Ne7-f5 Be2-g4 Nf5xd4 h5-h6) +0.22/25 1} 13. Nb5 {(Nc3-b5
Ra8-c8 c2-c4 Bb4-e7 Qh4-f4 d5xc4 Bf1xc4 a7-a6 Nb5-c3 Be7-d6 Qf4-e3 Nc6-e7
Nf3-e5 Bd6xe5) +0.02/20 3} Ba5 {(Bb4-a5 Nb5-c3 a7-a6 a2-a3 Ba5xc3 b2xc3
Nc6-e7 Nf3-e5 Bd7-b5 Kc1-b2 Bb5xf1 Rd1xf1 Ne7-f5 Qh4-h7 Qg7-f8 g2-g4 Nf5-d6
Qh7-h4 Qf8-e7 f2-f4 Qe7xh4 Rh1xh4 f7-f6) +0.22/21 1} 14. c4 {(c2-c4 a7-a6
Nb5-c3 Ba5xc3 b2xc3 d5xc4 Bf1xc4 Nc6-e7 Nf3-e5 Bd7-c6 Qh4-h7 Qg7xh7 Rh1xh7
Rg8xg2 Rh7xf7 Bc6-d5 Rd1-h1 O-O-O) +0.25/19 2} dxc4 {(d5xc4 d4-d5 e6xd5
Rd1xd5 Qg7-g4 Nb5-d4 Qg4xh4 Rh1xh4 O-O-O Nd4xc6 Bd7xc6 Rd5xa5 Bc6xf3 Ra5xa7
Rd8-d1+ Kc1-c2 Kc8-b8 Ra7-a3 Rd1xf1 Ra3xf3 Rg8xg2 Rf3xf7 b7-b5 a2-a4
Rg2xf2+ Rf7xf2 Rf1xf2+ Kc2-b1 c7-c6 a4xb5 c6xb5 Rh4-h5 Kb8-b7 Rh5xb5+
Kb7-c6 Rb5-h5 Rf2-g2 Kb1-c1 Kc6-d6 Rh5-h6+ Kd6-d5 Rh6-h5+ Kd5-e4 Rh5-h7
Ke4-d4) +0.08/26 8} 15. Bxc4 {(Bf1xc4 a7-a6 Nb5-c3 Ba5xc3 b2xc3 Nc6-e7
Nf3-e5 Ne7-f5 Qh4-h7 Qg7xh7 Rh1xh7 Nf5-d6 Bc4-b3 Rg8xg2 Rh7-h8+ Ke8-e7)
+0.36/19 2} Ne7 {(Nc6-e7 Nf3-e5 Bd7xb5 Bc4xb5+ c7-c6 Bb5-c4 Ba5-c7 Qh4-h5
Bc7xe5 d4xe5 Ne7-d5 Kc1-b1 O-O-O g2-g3 Nd5-b6 Bc4-b3 Rd8xd1+ Rh1xd1 Rg8-h8
Qh5-e2 Nb6-d5 a2-a3 Kc8-b8 Kb1-a1 Qg7-g5 Ka1-b1 Rh8-h2 Bb3-c2) +0.29/21}
16. Ne5 {(Nf3-e5 Bd7xb5 Bc4xb5+ c7-c6 Bb5-c4 Ba5-c7 Qh4-f4 O-O-O Qf4xf7
Qg7xf7 Ne5xf7 Rd8-f8 Bc4xe6+ Kc8-b8 Nf7-e5 Rg8xg2) +0.52/19 2} Bxb5
{(Bd7xb5 Bc4xb5+ c7-c6 Bb5-c4 Ba5-c7 Qh4-h5 Bc7xe5 d4xe5 Ne7-d5 g2-g3 O-O-O
Kc1-b1 Rg8-h8 Qh5-e2 Rh8xh1 Rd1xh1 Qg7-g6+ Kb1-a1 Nd5-b4 Bc4-b3 Qg6-d3
Qe2xd3 Rd8xd3 Rh1-h8+ Kc8-d7 Rh8-b8 b7-b5 Rb8-b7+ Kd7-e8 a2-a3 Rd3xb3 a3xb4
Rb3-f3 Rb7-b8+ Ke8-e7 Rb8-b7+ Ke7-f8 Ka1-b1 Rf3xf2 Rb7xa7 Rf2-g2) +0.36/24
3} 17. Bxb5+ {(Bc4xb5+ c7-c6 Bb5-c4 Ba5-c7 Qh4-h5 Bc7xe5 d4xe5 Ne7-d5
Bc4xd5 c6xd5 Kc1-b1 Ra8-c8 Rd1-c1 Rc8xc1+ Rh1xc1) +0.36/19 1} c6 {(c7-c6
Bb5-e2 Ba5-c7 Qh4-e4 O-O-O Rh1-h7 Qg7xg2 Qe4xg2 Rg8xg2 Rh7xf7 Ne7-f5 Be2-f1
Rg2-g8 Bf1-c4 Nf5xd4 Ne5-f3 Nd4-b5 Bc4xe6+ Kc8-b8 Rd1xd8+ Rg8xd8 Nf3-g5
Bc7-b6 Ng5-e4 Rd8-e8 Be6-f5 Re8-h8 Rf7-d7 Rh8-h1+ Rd7-d1 Rh1-h2 Bf5-e6
Nb5-d4 Be6-c4 Kb8-c7 Rd1-d3 Rh2-h1+ Rd3-d1) +0.25/25 3} 18. Bc4 {(Bb5-c4
Ba5-c7 Qh4-h5 Bc7xe5 d4xe5 Ne7-g6 Qh5-g5 Qg7xe5 Qg5-d2 Qe5-f4 Qd2xf4 Ng6xf4
g2-g3 Nf4-d5 Bc4xd5 c6xd5) +0.24/20 3} Bc7 {(Ba5-c7 Qh4-h5 Bc7xe5 d4xe5
Ne7-d5 Kc1-b1 O-O-O g2-g3 Rg8-h8 Qh5-e2 Rh8xh1 Rd1xh1 Qg7-g6+ Kb1-a1 Nd5-b4
Bc4-b3 Qg6-g5 a2-a3 Nb4-a6 Qe2-f3 Qg5xe5 Qf3xf7 Na6-c5 Bb3-c2 Nc5-d3 Bc2xd3
Rd8xd3 Qf7-f8+ Rd3-d8 Qf8-f7 Qe5-e4 Rh1-c1 Kc8-b8 Ka1-a2 Qe4-d5+ Ka2-a1
a7-a6 Qf7-g7 Kb8-a7) +0.46/26 2} 19. Qh5 {(Qh4-h5 Bc7xe5 d4xe5 Ne7-d5
Bc4xd5 c6xd5 Kc1-b1 O-O-O Rd1-c1+ Kc8-b8 g2-g3 Rg8-h8 Qh5-e2 Rh8xh1 Rc1xh1
Qg7-g6+) +0.22/20 2} Bxe5 {(Bc7xe5 d4xe5 Ne7-d5 Kc1-b1 O-O-O Bc4xd5 c6xd5
Qh5-h7 Qg7xe5 Qh7xf7 Rg8-f8 Qf7-e7 Rd8-e8 Qe7-b4 Kc8-b8 Kb1-a1 Qe5-g7 f2-f3
Re8-d8 Qb4-d2 d5-d4 Qd2-e2 Qg7-g5 Rh1-e1 Rf8-e8 Qe2-e5+ Qg5xe5 Re1xe5
Kb8-c7 Ka1-b1 Kc7-c6 Re5-h5 Kc6-b6 g2-g4 d4-d3 Rd1-d2 e6-e5) +0.13/27 3}
20. dxe5 {(d4xe5 Ne7-d5 Bc4xd5 c6xd5 Kc1-b1 O-O-O Rd1-c1+ Kc8-b8 g2-g3
Rg8-h8 Qh5-e2 Qg7-g6+ Kb1-a1 Rh8xh1 Rc1xh1 d5-d4 Rh1-d1) 0.00/19 1} Nd5
{(Ne7-d5 Bc4xd5 c6xd5 Kc1-b1 Ra8-c8 g2-g3 Ke8-d7 Qh5-h4 Qg7xe5 Qh4-a4+
Rc8-c6 Qa4xa7 Qe5-e4+ Kb1-a1 Qe4-b4 a2-a3 Qb4-b3 Qa7-d4 Rg8-a8 Rd1-d3
Qb3-c4 Qd4xc4 Rc6xc4 Rh1-h7 Kd7-e7 Ka1-a2 Ra8-f8 Rd3-b3 Rc4-c7 Rb3-b4
Ke7-f6 Ka2-b3 Kf6-g6 Rh7-h1 Rf8-c8 f2-f4 Kg6-g7) +0.20/29 3} 21. Bxd5
{(Bc4xd5 c6xd5 Kc1-b1 O-O-O Rd1-c1+ Kc8-b8 Qh5-h7 Qg7xh7+ Rh1xh7 Rg8xg2
Rh7xf7 d5-d4 Rc1-c7 Rg2-g1+ Kb1-c2 d4-d3+ Kc2-d2 Rg1-b1 Rc7xb7+ Kb8-a8
Rb7xa7+) -0.07/19 1} cxd5 {(c6xd5 Kc1-b1 Ra8-c8 g2-g3 Ke8-d7 Qh5-h4 Qg7-g4
Qh4-f6 Qg4-g6+ Qf6xg6 Rg8xg6 Rh1-h7 Kd7-e7 Rd1-d3 Rg6-g5 Rd3-f3 Rc8-f8
Rf3-e3 d5-d4 Re3-b3 Rg5xe5 Rb3xb7+ Ke7-f6 Kb1-c2 Re5-e2+ Kc2-d3 Re2xf2
Kd3xd4 Rf2-g2 Kd4-e3 Rg2xg3+ Ke3-f4 Rg3-g1 Rh7-h6+ Rg1-g6 Rh6xg6+ Kf6xg6
Rb7xa7 Rf8-c8 Ra7-b7 Rc8-c2 Kf4-e5) +0.17/28 2} 22. Kb1 {(Kc1-b1 O-O-O
Rd1-c1+ Kc8-b8 Qh5-h7 Qg7xh7+ Rh1xh7 Rd8-f8 g2-g3 Rg8-g5 f2-f4 Rg5xg3 f4-f5
e6xf5 Rc1-f1 Rg3-e3 Rf1xf5 d5-d4 Kb1-c2) -0.13/20 1} Rc8 {(Ra8-c8 g2-g3
Ke8-d7 Qh5-h4 Qg7-g4 Qh4-f6 Qg4-g6+ Qf6xg6 Rg8xg6 Rh1-h7 Kd7-e7 Rd1-d3
Rg6-g5 Rd3-f3 Rc8-f8 Rf3-e3 d5-d4 Re3-b3 Rg5xe5 Rb3xb7+ Ke7-f6 Kb1-c2
Re5-e2+ Kc2-d3 Re2xf2 Kd3xd4 Rf2-g2 Kd4-e3 Rg2xg3+ Ke3-f4 Rg3-g1 Rh7-h6+
Rg1-g6 Rh6xg6+ f7xg6 Rb7xa7 g6-g5+ Kf4-g4 Rf8-h8 Ra7-a3 Rh8-h2 b2-b3 Kf6-g6
Ra3-a7) +0.08/28 2} 23. Qe2 {(Qh5-e2 a7-a6 Rd1-c1 Rc8xc1+ Rh1xc1 Qg7-g6+
Kb1-a1 Qg6-g5 Rc1-c7 Qg5xg2 Rc7-c8+ Ke8-d7 Rc8xg8 Qg2xg8 Qe2-e3 Qg8-g1+
Qe3-c1 Qg1xc1+) -0.12/19 3} Rc5 {(Rc8-c5 g2-g3 Qg7-g6+ Kb1-a1 Ke8-d7 Qe2-d2
Rg8-c8 Qd2-b4 b7-b5 Qb4-a3 Qg6-e4 Qa3xa7+ Rc8-c7 Qa7-b8 Qe4xh1 Qb8xc7+
Rc5xc7 Rd1xh1 Rc7-c2 f2-f4 Rc2-g2 Ka1-b1 Rg2xg3 Rh1-h7 Kd7-e7 Rh7-h8 Rg3-g4
Rh8-b8 Rg4xf4 Rb8-b7+ Ke7-f8 Rb7xb5 Rf4-f2 Rb5-b3 Kf8-g7 Rb3-g3+ Kg7-h6)
+0.40/24 4} 24. Rc1 {(Rd1-c1 Rc5xc1+ Rh1xc1 Qg7-g6+ Kb1-a1 Ke8-f8 Qe2-e3
Kf8-g7 g2-g4 Qg6-e4 Qe3-g5+ Qe4-g6 Qg5-f4 Rg8-h8 Rc1-c7 Rh8-h1+ Rc7-c1
Rh1xc1+ Qf4xc1 Qg6xg4) +0.41/18 2} Qxg2 {(Qg7xg2 a2-a3) +0.66/21 1} 25. a3
{(a2-a3 Rc5xc1+ Rh1xc1 Ke8-f8 Qe2-b5 Qg2-e4+ Kb1-a2 b7-b6 Qb5-d7 Qe4xe5
Rc1-c8+ Kf8-g7 Rc8xg8+ Kg7xg8 Qd7xa7 d5-d4 Qa7xb6 Qe5-d5+ Qb6-b3 d4-d3
Qb3xd5) +0.32/18 2} a6 {(a7-a6 Rc1xc5 Qg2xh1+ Kb1-a2 Rg8-g1 Rc5-c8+ Ke8-e7
Rc8-c7+ Ke7-f8 Rc7xb7 Qh1-e4 Qe2xa6 Kf8-g7 Qa6-b6 Rg1-c1 Qb6-d8 Qe4-c4+
Rb7-b3 d5-d4 Qd8-f6+ Kg7-g8 Qf6-d8+ Kg8-g7) 0.00/29 7} 26. Rxc5 {(Rc1xc5
Qg2xh1+ Kb1-a2 Rg8-g1 Rc5-c8+ Ke8-d7 Qe2-c2 Rg1-a1+ Ka2-b3 Qh1-f3+ Rc8-c3
Qf3-d1 Rc3-c7+ Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+ Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+)
+0.20/19 4} Qxh1+ {(Qg2xh1+ Kb1-a2 Rg8-g1 Rc5-c8+ Ke8-e7 Rc8-c7+ Ke7-f8
Rc7xb7 Qh1-e4 Qe2xa6 Kf8-g7 Qa6-b6 Rg1-c1 Qb6-e3 Qe4-b1+ Ka2-b3 Qb1-c2+
Kb3-a2 Qc2-b1+) 0.00/30 1} 27. Ka2 {(Kb1-a2 Rg8-g1 Rc5-c8+ Ke8-d7 Qe2-c2
Rg1-a1+ Ka2-b3 Qh1-f3+ Rc8-c3 Qf3-d1 Rc3-c7+ Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+
Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+) +0.08/19 1} Rg1 {(Rg8-g1 Rc5-c8+ Ke8-e7
Rc8-c7+ Ke7-f8 Rc7xb7 Qh1-e4 Qe2xa6 Kf8-g7 Qa6-b6 Rg1-c1 Qb6-e3 Qe4-b1+
Ka2-b3 Qb1-c2+ Kb3-a2 Qc2-b1+) 0.00/34 2} 28. Rc8+ {(Rc5-c8+ Ke8-d7 Qe2-c2
Rg1-a1+ Ka2-b3 Qh1-f3+ Rc8-c3 Qf3-d1 Rc3-c7+ Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+
Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+) +0.02/19 1} Kd7 {(Ke8-d7 Qe2-c2 Rg1-a1+
Ka2-b3 Qh1-f3+ Rc8-c3 Qf3-d1 Rc3-c7+ Kd7-e8 Rc7-c8+ Ke8-e7 Rc8-c7+ Ke7-e8)
0.00/35 2} 29. Qc2 {(Qe2-c2 Rg1-a1+ Ka2-b3 Qh1-f3+ Rc8-c3 Qf3-d1 Rc3-c7+
Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+ Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+) +0.03/20 1}
Ra1+ {(Rg1-a1+ Ka2-b3 Qh1-f3+ Rc8-c3 Qf3-d1 Rc3-c7+ Kd7-e8 Rc7-c8+ Ke8-e7
Rc8-c7+ Ke7-e8) 0.00/38 2} 30. Kb3 {(Ka2-b3 Qh1-f3+ Rc8-c3 Qf3-d1 Rc3-c7+
Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+ Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+) +0.03/20} Qd1
{(Qh1-d1 Rc8-c7+ Kd7-e8 Rc7-c8+ Ke8-e7 Rc8-c7+ Ke7-e8) 0.00/40 3} 31. Rc7+
{(Rc8-c7+ Kd7-e8 Rc7-c8+ Ke8-d7 Rc8-c7+ Kd7-e8 Rc7-c8+ Ke8-d7) +0.01/21 2}
Ke8 {(Kd7-e8 Rc7-c8+ Ke8-e7 Rc8-c7+ Ke7-e8) 0.00/41 1} 32. Rc8+ {(Rc7-c8+
Ke8-d7 Rc8-c7+ Kd7-e8 Rc7-c8+ Ke8-d7) 0.00/21 1} Kd7 {(Ke8-d7 Rc8-c7+
Kd7-e8 Rc7-c8+ Ke8-d7) 0.00/42 2} 33. Rc7+ {(Rc8-c7+ Kd7-e8 Rc7-c8+ Ke8-d7)
0.00/20 1} Ke8 {(Kd7-e8 Rc7-c8+ Ke8-d7) 0.00/43 2} 34. Rc8+ {(Rc7-c8+
Ke8-d7) 0.00/20 1} Kd7 {(Ke8-d7) 0.00/43 1 3-fold repetition} 1/2-1/2[/pgn]

[/url]

[Daniel Shawul]

Again you refer to 'massive hardware'. But the hardware of AlphaZero was not really massive. Just different. AlphaZero needs fewer transistors to run at 3500 Elo than Stockfish needs to run at 3400 Elo, at 'standard' TC. It is just that the transistors have to be connected in an entirely different way.

That the AlphaZero hardware can do some things 100 times faster than equivalent (in terms of number of transistors) x64 CPUs, doesn't mean a thing. My rowing boat is infinitely faster than my Ferrari, for crossing the English Channel.

4-TPUs can do 180 TFLOPs, while 64 CPU cores can do 1 TFLOP; that is like a 180x hardware advantage. It doesn't matter whether we call 4-TPUs massive or not, the 180x advantage given to A0 is there.

I am not interested in cost comparsions as they are irrelevant for algorithmic comparsions.

Btw, here is that big X I refered to https://youtu.be/A3ekFcZ3KNw?t=1682

[noobpwnftw]

Again you refer to 'massive hardware'. But the hardware of AlphaZero was not really massive. Just different. AlphaZero needs fewer transistors to run at 3500 Elo than Stockfish needs to run at 3400 Elo, at 'standard' TC. It is just that the transistors have to be connected in an entirely different way.

That the AlphaZero hardware can do some things 100 times faster than equivalent (in terms of number of transistors) x64 CPUs, doesn't mean a thing. My rowing boat is infinitely faster than my Ferrari, for crossing the English Channel.

Not massive?
A "single machine" packing in 4 Gen.2 TPUs of which is 4 GPUs each, that's 16 GPUs total. Each of that one GPU is performing about a double of TFLOPs compared to a 1080TI.
A similar hardware like that would be a DGX-2, which costs like half a million last time I checked.

One could build a machine that's less than 1/10 of a million where "Stock"fish will yield +100 ELO over book-less SF8 64-threads.

On the CPU side to build a kind of decent chess-playing "single machine", the least you can get is some 8-way server, which usually has no less than 128 cores if not hyper-threaded. Do measurement there instead some one core to one GPU comparison, computers running on one core was at 90s.

Where is the breakthrough in domain specific implementation exactly?

[frankp]

Where is the breakthrough in domain specific implementation exactly?

No one tells how to play - it learns for itself - perhaps ?

[Albert Silver]

Sigh..wake me up when it is 2800 elo running on singe CPU core, which is what every other engine uses in rating lists. As far as I am concerned, it is still a 2100 elo engine there.
I see so many excited people giving a hardware advantage to LCzero, like CCLS does for instance uses a GPU for LCzero and single core CPU for the rest of the engines.

Well, to begin with, I remember when Rybka was the first engine to take advantage of the 64-bit environment when every standard OS was 32-bit. It had a big speed-up, and no one was able to do the same at first. Shredder 64-bit came out a couple of months later but with zero speedup. I don't recall people saying that it needed to run in a 32-bit environment like everyone else to be 'fair'.

The advantage you complain about is just sour grapes in my book. For one thing, if CCLS or whomever offer a GPU, then it is up to the authors to take advantage of it, not for the one who is able to, to learn to dumb down his machine for 'fairness'.

Leela is designed to use a GPU for best performance. it is inherent in its design. If it reaches 100 Elo better than everyone else on my computer because it alone can use the GPU to best advantage, while all others are weaker because they are only able to use the CPU, guess how much I (and everyone who analyzes with engines) will care?