cheng4 0.38 release

mar · Post by **mar** » Sun Jan 18, 2015 10:19 am

Hi all,
new cheng4 is available here http://vlasak.biz/cheng as usual.
Most of the gain (I expect some 30 real elo, maybe more) is due to eval tuning using "Texel tuning method" by Peter Osterlund (thanks).
Also thanks to Graham for providing book lines, Lars for testing and Emil for hosting.

cc2150dx · Post by **cc2150dx** » Sun Jan 18, 2015 10:45 am

I'll be testing this for sure.

Thanks for the update Martin

and Graham, Lars and Emil

Jason

Marek Soszynski · Post by **Marek Soszynski** » Sun Jan 18, 2015 10:59 am

The Linux compile, in particular, is much appreciated. Thank you!

Joerg Oster · Post by **Joerg Oster** » Sun Jan 18, 2015 12:28 pm

Great work. Thanks for sharing.

reflectionofpower · Post by **reflectionofpower** » Sun Jan 18, 2015 1:01 pm

Its runs slower than .36c speed wise in the nodes, it is very noticeable to me.
This isn't necessarily a bad thing, I was just wondering what you did to this current version?

mar · Post by **mar** » Sun Jan 18, 2015 1:06 pm

reflectionofpower wrote:Its runs slower than .36c speed wise in the nodes, it is very noticeable to me.
This isn't necessarily a bad thing, I was just wondering what you did to this current version?

Are you sure? I don't see any decrease in nps from the startposition between 0.36c and 0.38.
Could you start 0.36c and type go depth 20 and do the same for 0.38 and post nps when it finishes?

reflectionofpower · Post by **reflectionofpower** » Sun Jan 18, 2015 1:12 pm

mar wrote:
reflectionofpower wrote:Its runs slower than .36c speed wise in the nodes, it is very noticeable to me.
This isn't necessarily a bad thing, I was just wondering what you did to this current version?
Are you sure? I don't see any decrease in nps from the startposition between 0.36c and 0.38.
Could you start 0.36c and type go depth 20 and do the same for 0.38 and post nps when it finishes?

A SIGNIFICANT difference! newest version takes 40 seconds to reach 20 ply whereas older version takes 7 seconds! This is running 4 CPU's on an i7

I was also looking at the ply speed mainly so I should clarify mysef, node speed is the same but depth reached is much slower in new version.

mar · Post by **mar** » Sun Jan 18, 2015 1:15 pm

reflectionofpower wrote:A SIGNIFICANT difference! newest version takes 40 seconds to reach 20 ply whereas older version takes 7 seconds! This is running 4 CPU's on an i7

Ah yes so you mean nominal depth. Should be ok then. Also note that smp in nondeterministic.

Joerg Oster · Post by **Joerg Oster** » Sun Jan 18, 2015 2:42 pm

Just out of curiosity, I ran a quick 500 games match, fixed depth 12, Cheng4 1 Thread against Cheng4 8 Threads.

Code: Select all

Finished game 500 &#40;Cheng4-1T vs Cheng4-8T&#41;&#58; 1/2-1/2 &#123;Draw by 3-fold repetition&#125;
Score of Cheng4-8T vs Cheng4-1T&#58; 152 - 114 - 234  &#91;0.538&#93; 500
ELO difference&#58; 26
Finished match

Nice. I like it.

reflectionofpower · Post by **reflectionofpower** » Sun Jan 18, 2015 2:48 pm

Joerg Oster wrote:Just out of curiosity, I ran a quick 500 games match, fixed depth 12, Cheng4 1 Thread against Cheng4 8 Threads.
Code: Select all
Finished game 500 &#40;Cheng4-1T vs Cheng4-8T&#41;&#58; 1/2-1/2 &#123;Draw by 3-fold repetition&#125;
Score of Cheng4-8T vs Cheng4-1T&#58; 152 - 114 - 234  &#91;0.538&#93; 500
ELO difference&#58; 26
Finished match
Nice. I like it.

You could run a 1 thread verse a 48 thread, it makes no difference when running it into a depth limitation. They both reach 12 ply and then it's over. Try running it on 3 0 games and the 8 thread will cremate the solo act.

cheng4 0.38 release

cheng4 0.38 release

Re: cheng4 0.38 release

Re: cheng4 0.38 release

Re: cheng4 0.38 release

Re: cheng4 0.38 release

Re: cheng4 0.38 release

Re: cheng4 0.38 release

Re: cheng4 0.38 release

Re: cheng4 0.38 release

Re: cheng4 0.38 release