Zeta 0970 - Alpha II Release

[smatovic]

Heyho,

today is the 13th July (the montenegrin Independence Day) and i release the second Alpha Version of "Zeta", a Chess Engine coded in OpenCL, and therefore able to run on GPUs, CPUs or APUs.

Zeta uses an Selective Search Algorithm (Best-First) and includes a simple Evaluation Function based on Piece-Values and Piece-Square-Tables only, i guess its ELO is about 1600.

The new Version supports now Castle and EnPassant Moves, so it is comparable to other Engines.

Zeta runs on AMD/Intel CPUs and on Nvidia/AMD GPUs,
current GPU-Drivers already include OpenCL-Support,
to run on an CPU you will need to install an OpenCL SDK from your vendor.

On an Quad-Core with 2.2 GHz Zeta 0970 makes about 1.3 Mnps, on an AMD HD7750 GPU about 700 Knps (would be nice to know how it performs on an highend GPU).

Features
- Selective Search Algorithm
- Winboard/Xboard-Support
- OpenSource (GPL)
- Windows 64 Bit Binaries available
- Linux 64 Bit Binaries available

Changes
- Castle and EnPassant moves inside
- Repetition Detection inside

Missing
- 50-Move-Rule
- More precise Time-Control
- Multi-GPU Support
- auto-config

Link to Source (binaries are in bin folder):
https://github.com/smatovic/Zeta/tree/zeta_cl_0970

You will need to edit some values manually in the config.cfg file to run the engine.

--
Srdja

PS: check that donation button, Zeta needs new hardware: http://zeta-chess.blogspot.de/2012/06/d ... quest.html

[Ozymandias]

1600 is probably a little low for blunder check, but as soon as it gets over 2000 this could be a nice way to take advantage of SB's GPU. Keep it up!

[stevenaaus]

Code: Select all

zeta.c:25:21: fatal error: zobrist.h: No such file or directory

My Linux NV driver has installed libOpenCL, but the SDK is 145mb - too big.

[smatovic]

Code: Select all

zeta.c:25:21: fatal error: zobrist.h: No such file or directory

ahh thanks, i forgot to commit that one, i''ve just updated the repo.


--
Srdja

[gleperlier]

Code: Select all

zeta.c:25:21: fatal error: zobrist.h: No such file or directory

ahh thanks, i forgot to commit that one, i''ve just updated the repo.


--
Srdja

Just a noob question : How do you say an engine to run on the GPU instead of the CPU ?

Gab

[smatovic]

Just a noob question : How do you say an engine to run on the GPU instead of the CPU ?

in the file config.cfg you can specify which platform you want to use...if GPU or CPU...

--
Srdja

[gleperlier]

Just a noob question : How do you say an engine to run on the GPU instead of the CPU ?

in the file config.cfg you can specify which platform you want to use...if GPU or CPU...

--
Srdja

Great !

Gab

[stevenaaus]

Hmmm - it slows my box to a crawl
and i cant get it to work with Scid vs PC with Intel CPU or NV GPU.
Probably my old gpu is no use anyway.
NVIDIA GPU GeForce 8400 GS (G98) at PCI:1:0:0 (GPU-0)

[smatovic]

Hmmm - it slows my box to a crawl

Yepp, during GPU computation the box freezes

and i cant get it to work with Scid vs PC.

I ve tested only Xboard...will test it with Scid vs PC these days....

Probably my old gpu is no use anyway.
NVIDIA GPU GeForce 8400 GS (G98) at PCI:1:0:0 (GPU-0)

Yepp, 8400 is a bit too slow...it has only 16 Cores...

http://en.wikipedia.org/wiki/Comparison ... sing_units

I would suggest at least an NV 8800.
--
Srdja

[smatovic]

and i cant get it to work with Scid vs PC with Intel CPU or NV GPU.

Does it run on the command line? >zeta.bin >new >go

CPUs need some OpenCL-Driver, either Intel OpenCL SDK or AMD OpenCL SDK works for Intel CPUs SSE >= 4 too.

For the 8400 GPU i suggest following setup,assuming you have 512 MB RAM:


// Zeta OpenCL config file

threadsX: 1; // amount of SIMD units or CPU cores
threadsY: 4; // multi for warp
threadsZ: 32; // Warp or Wavefront Size

nodes_per_second: 25000; // nps of the OpenCL device, for time control
max_nodes: 0; // max nodes to search, if set to 0 time management will define this value
max_nodes_to_expand: 3355443; // max nodes to expand, each node needs 40 Bytes of RAM, 3355443 * 40 Bytes <= 128 MB, 6710886 * 40 Bytes <= 256 MB, 13421772 * 40 Bytes <= 512 MB
max_leaf_depth: 0; // search leaf nodes to depth, if set to 0 than only Qsearch will be done
max_depth: 43; // max internal depth
reuse_node_tree: 0; // experimental feature, 0==false || 1==true , computed node-tree will be prepared and reused for next iteration

opencl_platform_id: 0; // 0==NVIDIA GPU, 1==AMD CPU, 2==AMD GPU, 3==Intel CPU