Monte carlo on a NVIDIA GPU ?

Discussion of chess software programming and technical issues.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
p94100687
Posts: 3
Joined: Mon Feb 03, 2014 12:45 am
Location: Seattle
Contact:

Re: Monte carlo on a NVIDIA GPU ?

Post by p94100687 » Mon Feb 03, 2014 2:12 pm

Hello

I have a concept using FPGA boards (relatively cheeo, around 100-300 per board) that could do what you described i just hope i am not too late,

for contact details here my linkedin www.linkedin.com/pub/peter-paul-troendle/65/1a4/331/


i was myself thinking of making a chess game algorythm that can learn using the FPGA and the principle of concurrence

Hope to hear from you


PS: here people also think the way you do about levy etc.. i mean the marketing and other that walk on our shoes ;)

Peter
Forging a programming language for math, games and graphics

Edoardo Manino
Posts: 70
Joined: Wed Jul 11, 2012 10:43 pm
Location: Turin, Italy

Re: Monte carlo on a NVIDIA GPU ?

Post by Edoardo Manino » Mon Feb 03, 2014 4:15 pm

Hello Marco,
there's the zeta chess project
https://chessprogramming.wikispaces.com/Zeta
which uses OpenCL insted of CUDA
if I'm not wrong

syzygy
Posts: 4455
Joined: Tue Feb 28, 2012 10:56 pm

Re: Monte carlo on a NVIDIA GPU ?

Post by syzygy » Mon Feb 03, 2014 4:26 pm

Edoardo Manino wrote:Hello Marco,
there's the zeta chess project
https://chessprogramming.wikispaces.com/Zeta
which uses OpenCL insted of CUDA
if I'm not wrong
Note that you're responding to a 6-year old thread that has been revived by someone who seems to be mainly interested in advertising for himself.

jpqy
Posts: 522
Joined: Thu Apr 24, 2008 7:31 am
Location: Belgium

Re: Monte carlo on a NVIDIA GPU ?

Post by jpqy » Mon Feb 03, 2014 4:46 pm

When you do a search here in Talkchess you find zeta from Srdja Matovic

http://talkchess.com/forum/viewtopic.ph ... 60df5ead06

p94100687
Posts: 3
Joined: Mon Feb 03, 2014 12:45 am
Location: Seattle
Contact:

Re: Monte carlo on a NVIDIA GPU ?

Post by p94100687 » Mon Feb 03, 2014 10:03 pm

Hello

It is not about me that I am advertising myself, it is about an engine architecture on FPGA.

It does not yet exist.

Here to say that I am seeking to find ways how to developp it and wanted some ideas and in the same time inform Alba about that possibility

I am offering a connection in regards to some parallel LOGICS not processing, here look at FPGA and you will see what is meant

The comment you made is not really productive and I found it was not much a contribution, if I would be advertising please ask admin to delete my account, I am not I am replying to alba, that now all these years later the last years has seen NEW FPGA's with DSP blocks and RAM within the chip allowing very much that one cannot imagine

Another point here
https://github.com/smatovic/Zeta/tree/zeta_cl_0970
the link is not active and it is again not too productive honestly to give replyes that are not leading anywhere

I am looking to reply Marco Costalba very to the point question more not

Nice day to all

Peter
Forging a programming language for math, games and graphics

xmas79
Posts: 286
Joined: Mon Jun 03, 2013 5:05 pm
Location: Italy

Re: Monte carlo on a NVIDIA GPU ?

Post by xmas79 » Tue Feb 04, 2014 8:02 am

If I recall correctly, there was a time when ASICs like chessmachine worked like beasts, then came Hydra with its "assisted" FPGA boards and then dismissed in 2006 because not so much competitive. I think FPGAs cannot give a boost on chess development because search cannot be parallelized very well. A speed increase of 16x or 32x is very hard to reach with modern CPUs architectures, not because nowdays architectures are not well suited for the task, but because Alpha-Beta searches (the most effective by the way) are not too much scalable. So what you can do with 1024 processing logics that do the same thing? I suppose you wouldn't go that far....

And then came the GPUs...

Best regards,
Natale.

p94100687
Posts: 3
Joined: Mon Feb 03, 2014 12:45 am
Location: Seattle
Contact:

Re: Monte carlo on a NVIDIA GPU ?

Post by p94100687 » Tue Feb 04, 2014 8:54 am

Thank you,

The Nexus4 Xilinx newcomer board with Artix7, has 250 Multiply Adders and 2Mb+ RAM both inside the FPGA fabric. The working memory and operating blocks are all accessible within one clock cycle,

It is a terrible improvement the FPGA 's are undergoing the only issue is the human mind how to find how to exploit all these potentials

So here I set a goal which is clear to get the FPGA programmable so they can outperform the GPU and that is where this is totally new ground, pionneering and really hard

So I appreciate people helping proactively even with only a good word and I am looking at getting the algorythms into it basically if it works it is good so I am not a Marketing guy that does not care,

I hope to have more info's on how to join forces to make a FPGA version running and see with Stockfish perhaps Costalba would have his request fullfilled, this would be what I am working on
Forging a programming language for math, games and graphics

Post Reply