Zeta v099a, gpu chess update

[smatovic]

i finished my current run on Zeta, my experimental gpu chess engine.

https://github.com/smatovic/Zeta

Zeta v099a is a port of Zeta Dva v0305 (est. 2000 Elo on CCRL) to OpenCL,
the engine uses 64 threads per worker for move gen, move pick an eval on the
same node, so it is able to utilize a gpu with hundreds of workers via an
parallel alphabeta search (lazy smp derivat).

The actual conclusion of the current iteration is, that an simple engine, with
standard chess programming techniques, can be ported to OpenCL to run on a gpu,
but it would take more effort to make the engine competitive in terms of nps,
heuristics, and parallel scaling.

--
Srdja

[Werewolf]

Very exciting.

Is it a UCI engine and if not do you have plans to make it UCI?

[smatovic]

No, not UCI, but xboard/winboard CECPv2,

if the enigne reaches ever 3000+ Elo i will consider a UCI version

Code: Select all

Not supported Xboard commands:
analyze        // enter analyze mode
?              // move now
draw           // handle draw offers
hard/easy      // turn on/off pondering
hint           // give user a hint move
bk // book lines

--
Srdja

[smatovic]

maybe of interest....

scaling on starting position, depth 12, on Nv Titan X...

http://talkchess.com/forum/viewtopic.ph ... 59&t=63517

--
Srdja

[tttony]

Excelent!

I have an AMD 6950 (800,1250) the nps is around ~300k

Tested vs Skiull (CCRL 1500)
[pgn]
[Event "Computer chess game"]
[Site "HIGH-PC"]
[Date "2017.04.11"]
[Round "?"]
[White "Zeta 099a"]
[Black "Skiull 0.1 x64 popcnt"]
[Result "1/2-1/2"]
[BlackElo "1500"]
[ECO "A00"]
[Opening "Creepy Crawly (Basman) Opening"]
[Time "11:07:27"]
[Variation "Sicilian Variation, 2.Nf3 Nc6"]
[WhiteElo "2000"]
[TimeControl "60+1"]
[Termination "normal"]
[PlyCount "85"]
[WhiteType "program"]
[BlackType "program"]

1. Nc3 {(b1c3) +0.59/1 3} d5 {(d7-d5) 0.00/7 2} 2. Nf3 {(g1f3 b8c6 d2d3
g8f6 g2g3) +0.30/5 1} d4 {(d5-d4) 0.00/7 2} 3. Nb5 {(c3b5 b8c6 e2e3 e7e5
f1d3) +0.63/5 2} Nc6 {(Nb8-c6) +0.25/7 2} 4. e3 {(e2e3 d4e3 d2e3 g8f6)
+0.08/4 2} e5 {(e7-e5) +0.20/7 2} 5. exd4 {(e3d4 e5d4 f1d3) +0.64/3 1} exd4
{(e5xd4) +0.20/7 2} 6. c3 {(c2c3 d4d3 g2g3) +0.63/3 0} dxc3 {(d4xc3)
+0.15/7 2} 7. dxc3 {(d2c3 g8f6 c1e3 g7g6) +0.21/5 1} Nf6 {(Ng8-f6) -0.10/6
2} 8. Bd3 {(f1d3 d8e7 c1e3 f6d5) +0.28/4 2} Bg4 {(Bc8-g4) +0.15/6 2} 9. h3
{(h2h3 c6e5 b5c7 d8c7) +0.35/4 1} Bxf3 {(Bg4xf3) +0.30/7 2} 10. Qxf3 {(d1f3
c6e5 f3e3 d8e7 d3f5) +0.28/5 1} a6 {(a7-a6) +0.20/6 2} 11. Qe3+ {(f3e3 f8e7
b5d4 c6d4 e3d4) +0.21/5 1} Be7 {(Bf8-e7) +0.35/7 2} 12. Nd4 {(b5d4 c6d4
e3d4 d8d4 c3d4 f6d5) -0.01/6 1} Nxd4 {(Nc6xd4) +0.20/7 2} 13. Qxd4 {(e3d4
d8d4 c3d4 h7h6 c1f4 f6d5) +0.09/6 1} Qxd4 {(Qd8xd4) +0.90/7 2} 14. cxd4
{(c3d4 g7g6 c1e3 f6d5 a2a3 c7c6) -0.05/6 3} Rd8 {(Ra8-d8) +1.05/7 2} 15.
Be3 {(c1e3 c7c5 f2f3 c5d4 e3g5 f6d5 g5d2) -0.99/7 3} c5 {(c7-c5) +1.10/7 2}
16. f3 {(f2f3 c5d4 e3d2 g7g6 d2g5 f6d5) -0.97/6 1} cxd4 {(c5xd4) +1.45/8 2}
17. Bd2 {(e3d2 f6d5 e1c1 g7g6 h1e1 b7b5) -1.04/6 1} O-O {(O-O) +1.55/7 1}
18. Rd1 {(a1d1 f6d5 a2a3 e7h4 e1f1 d5e3) -1.05/6 1} Rd6 {(Rd8-d6) +1.55/7
1} 19. a3 {(a2a3 f6d5 b2b4 d6e6 e1f2) -0.83/5 1} Rb6 {(Rd6-b6) +1.70/7 1}
20. b4 {(b2b4 f6d5 g2g4 b6e6 e1f2 e7h4) -0.81/6 1} Re6+ {(Rb6-e6+) +1.70/7
1} 21. Kf1 {(e1f1 f6d5 g2g4 f7f6 d1e1) -0.78/5 1} Nd5 {(Nf6-d5) +1.60/6 1}
22. g4 {(g2g4 f7f6 f1f2 h7h6 h1e1 e6e1) -0.83/6 2} Rd8 {(Rf8-d8) +1.75/6 1}
23. Re1 {(d1e1 e6e1 f1e1 h7h6 h1f1) -0.83/5 1} Rxe1+ {(Re6xe1+) +1.90/7 1}
24. Kxe1 {(f1e1 e7h4 e1e2 g7g6 h1c1) -0.98/5 1} Bf6 {(Be7-f6) +1.80/7 1}
25. Kf2 {(e1f2 g7g6 h1d1 b7b5 d2h6) -0.98/6 2} h6 {(h7-h6) +1.75/6 1} 26.
Rc1 {(h1c1 f6e5 c1e1) -0.97/5 2} Be5 {(Bf6-e5) +1.65/6 1} 27. Rc5 {(c1c5
e5f4 d3e4 f4d2 c5d5 d2e3) -0.87/6 1} Bf4 {(Be5-f4) +1.65/7 1} 28. Be4
{(d3e4 f4d2 c5d5 d2e3 f2e2 d8d5) -0.87/6 1} Bxd2 {(Bf4xd2) +1.40/8 1} 29.
Rxd5 {(c5d5 d8d5 e4d5 d2c1 a3a4 c1e3 f2e2) -0.90/7 1} Rxd5 {(Rd8xd5)
+1.30/8 1} 30. Bxd5 {(e4d5 b7b5 d5b7 d2c1 b7a6 c1a3 a6b5) -0.98/7 1} b5
{(b7-b5) +1.30/9 1} 31. Bb7 {(d5b7 d2c1 b7a6 c1a3 a6b5 a3b4 b5d3 f7f6)
-0.95/8 2} Bc1 {(Bd2-c1) +1.10/10 1} 32. Bxa6 {(b7a6 c1a3 f2e2 a3b4 e2d3
b4d6 a6b5 d6e5) -0.93/8 1} Bxa3 {(Bc1xa3) +1.10/10 1} 33. Ke2 {(f2e2 a3b4
e2d3 b4c3 a6b5 f7f6 b5d7 g7g5 d7f5) -1.02/9 2} Bxb4 {(Ba3xb4) +1.20/10 1}
34. Kd3 {(e2d3 b4c3 a6b5 f7f6 b5d7 g7g5 d7f5 g8f7) -1.04/8 1} Bc5 {(Bb4-c5)
+1.20/9 1} 35. Bxb5 {(a6b5 f7f6 d3c4 c5b6 c4d5 g7g5 b5d3 g8f7) -1.03/8 1}
Kf8 {(Kg8-f8) +1.15/9 1} 36. Ke4 {(d3e4 c5b6 b5d3 f7f6 e4f5 b6c5 f5g6 c5d6)
-0.90/8 1} Ke7 {(Kf8-e7) +1.10/9 1} 37. Kd5 {(e4d5 c5a7 b5a6 f7f6 a6d3 g7g5
d5c6 a7b8) -0.84/8 1} Ba7 {(Bc5-a7) +1.15/9 1} 38. Bd3 {(b5d3 f7f6 d3f5
a7b6 d5c6 b6d8 f5d3 g7g5) -0.89/8 1} g5 {(g7-g5) +1.05/10 1} 39. Kc6 {(d5c6
e7f8 d3f5 f7f6 c6b5 f8e7 f5d3 e7d6) -0.86/8 1} Ke6 {(Ke7-e6) +1.25/9 1} 40.
Bc4+ {(d3c4 e6f6 c6d6 a7b6 d6d5 f6g6 c4d3 g6f6) -0.86/8 2} Ke7 {(Ke6-e7)
+1.25/9 1} 41. Bd3 {(c4d3 e7f8 d3c4 f7f6 c6d5 a7b6 d5c6 b6d8 c4d3) -0.90/9
1} Ke6 {(Ke7-e6) +1.25/9 1} 42. Bc4+ {(d3c4 e6f6 c6d6 a7b6 d6d5 f6g6 c4d3
g6f6) -0.87/8 2} Ke7 {(Ke6-e7) +1.25/9 1} 43. Bd3 {(c4d3 f7f6 c6d5 a7b6
d3e4 e7d7 e4d3 d7c8 d5e4) -1.05/9 1 3-fold repetition} 1/2-1/2
[/pgn]

[smatovic]

Excelent!

Thanks, still a lot of work to be done...

I have an AMD 6950 (800,1250) the nps is around ~300k
Tested vs Skiull (CCRL 1500)

Cool, nice to see it running,
i chose portability over speed,
so it should be able to run on the very first gpus
with OpenCL 1.x support (>=2008).

If you are interested, you can run an smp benchmark for your device,
start the engine in command line and enter following sequence

new
sd 12
st 2000
benchsmp

but it will take some minutes to finish,
and some gpu timeout may occure during run.

--
Srdja

[jpqy]

When i dubble click zeta099aw64.exe

I get this:
[2017-04-13 14:48:46] Error: In clReleaseMemObject (GLOBAL_BOARD_Buffer)
[2017-04-13 14:48:46] status:-38

JP.

[smatovic]

When i dubble click zeta099aw64.exe

I get this:
[2017-04-13 14:48:46] Error: In clReleaseMemObject (GLOBAL_BOARD_Buffer)
[2017-04-13 14:48:46] status:-38

it is still an alpha version, so it is not user friendly to get it to run,
did the guessconfig command process?

Usage:
First make sure you have an working OpenCL Runtime Environment,
start the zeta executable in command line with -dl option to list
all available OpenCL devices on host

Second check the OpenCL device and create a config.ini file for the engine,
zeta -p 0 -d 0 --guessconfigx
Where p is the selected platform id and d is the selected device id.

Third rename the created config file to config.ini and start the engine.

WARNING:
It is recommended to run the engine on an discrete GPU,
without display connected,
otherwise system and display can freeze or crash during computation.

Some GPU drivers have an timeout of 5 seconds if GPU is connected to display.
So make sure to use an discrete GPU or set proper time controls.

Windows OS have an internal gpu timeout, double click the .reg file
"SetWindowsGPUTimeoutTo20s.reg"
and reboot the OS to set the timeout to 20 seconds.

--
Srdja

[MikeB]

cool - keep it up...