Author: vladan
Date: 03:04:29 10/27/05
Go up one level in this thread
On October 27, 2005 at 05:09:11, Dagh Nielsen wrote: >Sedat has a list of Axon Benchmark results here: > >http://www.geocities.com/sedatchess/hardware.html > >My problem is that I am not sure how those benchmarks corresponds to the speed >with which real chess engines will run on different hardware. And I would like >to know this so I can see how much I would benefit from a hardware upgrade :-) > >Take a look at these three benchmarks: > >AMD Athlon 64 FX-55 2.60 GHz (00:00:24.4 sec.) [3774488 pps.] PowX=11.759 >AMD Athlon 64 3800+ 2.40 GHz (00:00:25.8 sec.) [3569671 pps.] PowX=11.121 >AMD Athlon XP 2400+ 2.00 GHz (00:00:35.6 sec.) [2587008 pps.] PowX=8.060 > >I have the last one, with a PowX at 8000. I am a little surprised that the >FX-55, according to Axon Benchmark, is less than 50% faster at calculating chess >positions than my old low end computer. I think I bought my computer about 2 >years ago, cheaply. > >So my question is, if I upgraded to an AMD Athlon 64 3800+ or such, would my >Fruit only calculate about 40% more kn/s? In other words, do the Axon Benchmarks >correspond linearly to performance in real chess engines? (If not, why! One >issue I'm wondering about is 32 bit vs. 64 bit.) > >I hope that I can get some help comparing the performance of my fruit 2.2 to the >performance on some faster computers. Here is Fruit 2.2 calculation of the >starting position on my AMD Athlon XP 2400+, done in Arena. I used 256MB for >hashtables, and made sure to clear them before starting calculation. > >[D] rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 > >Fruit 2.2: > 1/1 00:00 3 0 +0,34 Nc3 > 2/2 00:00 44 0 +0,20 Nc3 Nc6 > 3/3 00:00 148 0 +0,34 Nc3 Nc6 Nf3 > 4/6 00:00 668 0 +0,20 Nc3 Nc6 Nf3 Nf6 > 5/9 00:00 2.389 0 +0,28 Nc3 Nc6 Nf3 Nf6 d4 > 6/12 00:00 5.848 0 +0,20 Nc3 Nc6 Nf3 Nf6 d4 d5 > 7/13 00:00 15.807 0 +0,22 Nc3 Nc6 Nf3 Nf6 d4 d5 Bf4 > 8/16 00:00 33.201 0 +0,20 Nc3 Nc6 Nf3 Nf6 d4 d5 Bf4 Bf5 > 9/19 00:00 98.842 0 +0,06 Nc3 Nf6 Nf3 Nc6 d4 d5 Qd3 Qd6 Bg5 > 10/20 00:01 209.821 0 +0,26 Nc3 Nf6 Nf3 Nc6 d4 d5 Bf4 Bf5 Nb5 Rc8 > 11/26 00:01 391.292 0 +0,15 Nc3 Nf6 Nf3 Nc6 d4 d5 e3 Bf5 Bd3 Bxd3 Qxd3 e6 > 12/28 00:02 936.216 565.174 +0,20 Nc3 Nf6 Nf3 d5 d4 e6 e3 Bd6 Bd3 Nc6 0-0 >0-0 > 13/44 00:04 2.345.522 575.000 +0,12 Nc3 Nf6 Nf3 d5 d4 e6 e3 Bd6 Bd3 Nc6 0-0 >0-0 Ng5 > 14/44 00:10 5.878.749 578.075 +0,27 Nc3 Nf6 Nf3 Nc6 d4 e6 e3 Bb4 Bd3 Nd5 Bd2 >Bxc3 bxc3 0-0 0-0 > 14/44 00:16 9.163.824 576.052 +0,28 Nf3 Nf6 d4 Nc6 e3 d5 Nc3 Bd7 Bd3 Nb4 0-0 >h5 Be2 h4 > 15/44 00:29 16.567.226 578.268 +0,12 Nf3 Nf6 Nc3 d5 d4 Nc6 Bf4 Nh5 Bd2 Bf5 e3 >e6 Bb5 Bd6 0-0 > 15/44 00:38 21.889.762 579.769 +0,14 Nc3 Nf6 e4 Nc6 Nf3 e5 Bb5 Bb4 Bxc6 dxc6 >Nxe5 0-0 0-0 Re8 d4 Bxc3 bxc3 Nxe4 > 15/44 01:19 46.212.934 584.218 +0,21 e4 e5 Nf3 Nc6 d4 exd4 Nxd4 Bc5 Nxc6 Qf6 >f4 Qxc6 Nc3 Qb6 Bc4 Bf2+ Kf1 > 16/44 01:58 68.376.898 581.453 +0,25 e4 e5 Nf3 Nc6 Nc3 Nf6 d4 Bb4 Nxe5 Nxe4 >Qf3 Nf6 Be3 0-0 Bd3 Nxe5 dxe5 > 17/44 03:20 115.570.482 578.326 +0,19 e4 e5 Nf3 Nc6 Nc3 Nf6 d4 exd4 Nxd4 Bb4 >Nxc6 bxc6 Qd4 Qe7 Bd3 Bc5 Qa4 0-0 0-0 > >I would be very curious to see how fast other Fruit 2.2 users can reach depth >17, so we can compare performance. > >Thanks in advance. Dear mr. Nielsen, Thank you for using Axon Benchmark. Mr. Sedat was very kind to collect and present the most completed Axon Benchmark list on his site. For benchmarking, the program uses my embedded Axon v1.0 chess engine, and analyzes one special middlegame position. The engine is totally written in assembly language, it is the mixture of 16 and 32 bit machine instructions. It does not use MMX extensions, 64 instructions or other special processor features (instruction sets, logic processors etc.). Only one standard processor in system (single CPU, primary and secondary hash) is the test object. These facts determine its compatibility. It is very useful to test and compare chess engines running only on single processors which have similar software structure. Also it measures very precisely processor architecture advances connected with chess engines. As you noticed FX 2600 MHz machine has 50% better chess performances using only about 33% higher clock frequency. Better architecture. And that’s it. If you intend to use multiprocessor engines (like Fritz or Shredder) on multicore or multithreading processors you must use their own benchmarks. The method of using start position for benchmarking is a little obsolete. If you do it in that way, it is much better to download some standard EPD tests. Almost all modern interfaces (Shredder Classic, Arena, ChessBase …) have the option to analyze EPDs with different engines. Best regards, Vladan Vuckovic, the author of Axon Benchmark.
This page took 0.02 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.