BikJump 2.0 coming soon!
4 posters
Page 1 of 1
BikJump 2.0 coming soon!
I am enthusiatisc! The last two weeks I have spent a lot of my free time completely rewriting the first generation BikJump (v1.x) into the second generation (v2.x) by replacing the mailbox representation with a home-brewed bitboard representation, move generator, and evaluation function, while carefully revisiting and improving the search and other parts of the engine as well. The initial results are very promising. On a typical middle game position, the nodes-per-second rating of the true search increased from around 500Kns (32-bit and 64-bit) to 1300Kns (32-bit) and 1800Kns (64-bit). A lot more debugging, tuning, and experimentation is required, but I hope to release BikJump v2.0 really soon!
sounds promosing indeed!
Hi Aart Jan,
This sounds spectacular, 1800 Knodes/sec! On what hardware have you performed the test? I'm looking forward to the release.
Best wishes,
Stephan
This sounds spectacular, 1800 Knodes/sec! On what hardware have you performed the test? I'm looking forward to the release.
Best wishes,
Stephan
Stephan Vermeire (Brutus)- Number of posts : 32
Registration date : 2008-05-21
Re: BikJump 2.0 coming soon!
This was measured on a Intel Core 2 Quad Q6700 2.66GHz (8MB L2 Cache, FSB 1066MHz, 6GB RAM, 64-bit Vista Service Pack 1), still single threaded, using my favorite compiler (see http://www.intelcompiler.com :-) Of course, absolute numbers do not say everything, as all engines seem to count nodes differently (BikJump also counts nodes in quiescence, for example). The relative increase says more, however, as both BikJumps count nodes the same. I measured "perft" performance of the move generator at 14MNs, including incremental updates of state such as zobrist keys, which, for BikJump, translates to almost 2MNs effective search. And of course, all these numbers say nothing until BikJump actually performs well in practice.....
Re: BikJump 2.0 coming soon!
Aart J.C. Bik wrote:This was measured on a Intel Core 2 Quad Q6700 2.66GHz (8MB L2 Cache, FSB 1066MHz, 6GB RAM, 64-bit Vista Service Pack 1), still single threaded, using my favorite compiler (see http://www.intelcompiler.com :-) Of course, absolute numbers do not say everything, as all engines seem to count nodes differently (BikJump also counts nodes in quiescence, for example). The relative increase says more, however, as both BikJumps count nodes the same. I measured "perft" performance of the move generator at 14MNs, including incremental updates of state such as zobrist keys, which, for BikJump, translates to almost 2MNs effective search. And of course, all these numbers say nothing until BikJump actually performs well in practice.....
I downloaded the Intel compiler to try out the latest version.
It does not show up in my MS VC++ IDE. I saw that someone else has a similar problem (under 64 bit Windows Server 2008).
I tried to leave a message on the Intel message board to explain the problem, but it uses some sort of interface that my security setting short circuit so that I cannot post.
Re: BikJump 2.0 coming soon!
dann corbit wrote:It does not show up in my MS VC++ IDE. I saw that someone else has a similar problem (under 64 bit Windows Server 2008).
Make sure to install the integration in Visual Studio as well (last item on the list). The install sometimes takes a very long time to pop up the install wizard (so that one could think nothing is happening). Assuming you downloaded the compiler, simply try again from the Intel/Download/setup.exe and just install that. Hope this helps.
Page 1 of 1
Permissions in this forum:
You cannot reply to topics in this forum