Search found 13 matches

by vvarkey
Mon Feb 12, 2018 3:31 pm
Forum: Computer Chess Club: General Topics
Topic: Cloud TPUs now available for hire
Replies: 1
Views: 682

Cloud TPUs now available for hire

https://cloudplatform.googleblog.com/20 ... -beta.html
usage is billed by the second at the rate of $6.50 USD / Cloud TPU / hour
by vvarkey
Mon Dec 18, 2017 4:44 pm
Forum: Computer Chess Club: General Topics
Topic: I can't believe that so many people don't get it!
Replies: 149
Views: 33163

Re: I can't believe that so many people don't get it!

There is evidence that they keep a learn file with wins, losses, draws and a percentage chance of winning.
where is this evidence?
by vvarkey
Mon Dec 18, 2017 12:47 pm
Forum: Computer Chess Club: General Topics
Topic: In chess,AlphaZero outperformed Stockfish after just 4 hours
Replies: 64
Views: 13615

Re: In chess,AlphaZero outperformed Stockfish after just 4 h

And yet they trained against the most common positions in a human database and they did it 100 games per position and they did it against SF. 100 games with most common positions from human database were matches just like the main match. this was after all the training was complete. training was on...
by vvarkey
Mon Dec 18, 2017 6:32 am
Forum: Computer Chess Club: General Topics
Topic: AlphaGo and Stockfish played on similar hardware
Replies: 21
Views: 4962

Re: AlphaGo and Stockfish played on similar hardware

We can't test it but still there are some pretty good indications. When playing from the chess opening position A0 was 100Elo stronger than SF. When playing from Sicilian B40 (the opening that A0 trained very little) which was only 2 moves deep, A0 advantage drop to 38.5Elo. Huge drop in performanc...
by vvarkey
Mon Dec 18, 2017 5:19 am
Forum: Computer Chess Club: General Topics
Topic: AlphaGo and Stockfish played on similar hardware
Replies: 21
Views: 4962

Re: AlphaGo and Stockfish played on similar hardware

Ignore all the training that went into AlphaZero for a second. You really cannot do that, not even for a millisecond. The training (did I read 44 million games?) is worth more than 1,000 elo and probably much more. The learning was guided by NN to learn on the most promising lines thus narrowing th...
by vvarkey
Sun Dec 17, 2017 7:56 pm
Forum: Computer Chess Club: General Topics
Topic: AlphaGo and Stockfish played on similar hardware
Replies: 21
Views: 4962

Re: AlphaGo and Stockfish played on similar hardware

I am not sure that with 'contemporary CPUs' they mean 'single cores'. oops. in the actual TPU paper https://arxiv.org/pdf/1704.04760.pdf : The traditional CPU server is represented by an 18-core, dual-socket Haswell processor from Intel. The GPU accelerator is the Nvidia K80. So 4 TPUs = 18x15x4 = ...
by vvarkey
Sun Dec 17, 2017 6:12 pm
Forum: Computer Chess Club: General Topics
Topic: AlphaGo and Stockfish played on similar hardware
Replies: 21
Views: 4962

AlphaGo and Stockfish played on similar hardware

Ignore all the training that went into AlphaZero for a second. Per the paper, for the 100 games: AlphaZero and the previous AlphaGo Zero used a single machine with 4 TPUs. Stockfish and Elmo played at their strongest skill level using 64 threads and a hash size of 1GB. According to https://cloud.goo...
by vvarkey
Fri May 21, 2010 1:16 pm
Forum: Computer Chess Club: General Topics
Topic: IBM super-computer with 8792 processors!
Replies: 38
Views: 6088

Re: IBM super-computer with 8792 processors!

haha topalov the moral winner. that really made my day. I'm glad anand had the last laugh after the whole volcanic ash incident and the response from the organizers. To quote ( http://letters.chessdom.com/bcf-statement-anand-topalov ): ".....The organizers and the journalists in Bulgaria were bitter...
by vvarkey
Thu May 20, 2010 4:10 pm
Forum: Computer Chess Club: General Topics
Topic: Is that our Mark Lefler?
Replies: 15
Views: 3524

Re: Is that our Mark Lefler?

what's all this about rybka4 release being delayed so topalov had exclusive use of it??
"...Once we found out that he had some super computer, some cluster version of Rybka, something that can scale very well, it was obvious to us why Rybka 4 was being held back."