Andscacs - Test version 74024

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

User avatar
cdani
Posts: 2204
Joined: Sat Jan 18, 2014 10:24 am
Location: Andorra

Re: Andscacs - Test version 74024

Post by cdani »

Graham Banks wrote: Looks like I started my 8 core Swiss a few days too early. :wink:
I tried hard to have it for your tournament, but it was simply impossible. Anyway at the start I thought that to make it MP will be very hard, and in fact it was very easy :-) of course thanks to lazy SMP.

I forgot to mention that the problems with Chessbase where reported by Jose Velasco, and the ones with ponder by Sedat. Thanks!
User avatar
Graham Banks
Posts: 46742
Joined: Sun Feb 26, 2006 10:52 am
Location: Auckland, NZ

Re: Andscacs - Test version 74024

Post by Graham Banks »

cdani wrote:
Graham Banks wrote: Looks like I started my 8 core Swiss a few days too early. :wink:
I tried hard to have it for your tournament, but it was simply impossible. Anyway at the start I thought that to make it MP will be very hard, and in fact it was very easy :-) of course thanks to lazy SMP.

I forgot to mention that the problems with Chessbase where reported by Jose Velasco, and the ones with ponder by Sedat. Thanks!
No problem Daniel. There will be others. :)
Well done on your successful implementation of smp.
gbanksnz at gmail.com
cetormenter
Posts: 170
Joined: Sun Oct 28, 2012 9:46 pm

Re: Andscacs - Test version 74024

Post by cetormenter »

cdani wrote:Nice! Thanks. Very good your new version!! How many elo it wins in two threads, if you know?
Nirvana gains somewhere around 60 to 70 elo in self play going from 1 to 2 cores.
User avatar
cdani
Posts: 2204
Joined: Sat Jan 18, 2014 10:24 am
Location: Andorra

Re: Andscacs - Test version 74024

Post by cdani »

cetormenter wrote:
cdani wrote:Nice! Thanks. Very good your new version!! How many elo it wins in two threads, if you know?
Nirvana gains somewhere around 60 to 70 elo in self play going from 1 to 2 cores.
Aha.
In Andscacs, there are shared between threads the following tables:
* Hash
* Double Move Refutation
* Triple Move Refutation
* Four Move Refutation
* Piece hash
* Countermoves

May be all this other than the standard hash helps a little.
Jamal Bubker
Posts: 328
Joined: Mon May 24, 2010 4:32 pm

Re: Andscacs - Test version 74024

Post by Jamal Bubker »

Thank you Jose :D
User avatar
cdani
Posts: 2204
Joined: Sat Jan 18, 2014 10:24 am
Location: Andorra

Re: Andscacs - Test version 74024

Post by cdani »

cdani wrote: Anyway is not a lot stronger, may be 30 elo and I'm not sure after the addition of MP support. May be it has eaten some of the gains. I'm retesting it again now.
I tested again with one thread and seems to be more or less equal. So the overhead of passing a lot of global variables to pointed structures seems to be minimal.
cdani wrote:Now I'm testing with 8 threads, but it does not looks very good. I will report. May be it does not scale well.
Updated results:

2 threads: +61
4 threads: +108
6 threads: +117
7 threads: +119
8 threads: +96

Not good results with more than 4 threads. I will try with different variants. If I succeed with something, I will explain it.
User avatar
cdani
Posts: 2204
Joined: Sat Jan 18, 2014 10:24 am
Location: Andorra

Re: Andscacs - Test version 74024

Post by cdani »

cdani wrote: Updated results:

2 threads: +61
4 threads: +108
6 threads: +117
7 threads: +119
8 threads: +96

Not good results with more than 4 threads. I will try with different variants. If I succeed with something, I will explain it.
I achieved some improvement:

8 threads: +120

I explain it better on the programming section.
User avatar
cdani
Posts: 2204
Joined: Sat Jan 18, 2014 10:24 am
Location: Andorra

Re: Andscacs - Test version 74024

Post by cdani »

I obtained some nice improvements. I will stop posting about lazy eval and improvements in this thread, because doing it in the programming section is enough.

But I wanted to say that I updated the same zip file with the latest improved developing version, and I will continue to update it if I do improvements with lazy eval, but only the x64 popcnt version:

http://www.andscacs.com/andscacs074024.zip