Computer Chess Club Archives


Search

Terms

Messages

Subject: Better speak in THz (TeraHertz) terms!

Author: Omid David Tabibi

Date: 10:21:49 06/07/03

Go up one level in this thread


On June 07, 2003 at 11:54:30, emerson tan wrote:

>How many Ghz before we can get average 20ply per move on a tournament time
>control. I always want a computer that can see 10 moves ahead

Let's look at Crafty's analysis of the following position (on 733MHz, 64MB):

[D]r1b2rk1/1pq1bppp/p1nppn2/8/P2NPP2/2N1B3/1PP1B1PP/R2Q1RK1 w - - 0 1

Analysis by Crafty 19.03:

11.Bc4
  ²  (0.69)   Depth: 1/13   00:00:00
11.Bc4 Bd7
  ²  (0.58)   Depth: 2/13   00:00:00
11.Nxc6 bxc6
  ²  (0.68)   Depth: 2/13   00:00:00
11.Nxc6 bxc6 12.Bc4
  ±  (0.74)   Depth: 3/13   00:00:00
11.Nxc6 bxc6 12.b4 d5
  ²  (0.59)   Depth: 4/14   00:00:00
11.Nxc6 Qxc6 12.Qd4 d5 13.exd5 Nxd5 14.Nxd5 Qxd5 15.Qxd5 exd5
  ²  (0.62)   Depth: 5/20   00:00:00
11.Nxc6 bxc6 12.b4 d5 13.Qd4 Bb7
  ²  (0.54)   Depth: 6/20   00:00:00  63kN
11.Nxc6 bxc6 12.Qd2 a5 13.Rad1 Ba6 14.f5 Bxe2 15.Qxe2
  ²  (0.58)   Depth: 7/25   00:00:01  216kN
11.Nxc6 bxc6 12.Qd2 Bb7 13.b4 c5 14.b5 Rab8 15.bxa6 Nxe4
  ²  (0.49)   Depth: 8/25   00:00:03  611kN
11.Qd2 Bd7 12.Nb3 Nb4 13.Rad1 h6 14.f5 b6
  ²  (0.51)   Depth: 8/25   00:00:05  1195kN
11.Qd2 Bd7 12.Nb3 b6 13.Rad1 h6 14.Rfe1 Rfc8 15.Bc4
  ²  (0.52)   Depth: 9/30   00:00:09  2408kN
11.Qd2 Bd7 12.Nb3 b6 13.Bf3 h6 14.Rad1
  ²  (0.43)   Depth: 10/30   00:00:25  7242kN
11.Nxc6 bxc6 12.Qd3 a5 13.Rad1 Ba6 14.Qd2 Bxe2 15.Qxe2 d5 16.exd5 exd5
  ²  (0.44)   Depth: 10/30   00:00:37  10909kN
11.Nxc6 bxc6 12.Rc1 c5 13.Bf3 Bb7 14.Qe2 d5 15.exd5 exd5 16.Rcd1 Rad8
  ²  (0.41)   Depth: 11/32   00:01:29  28187kN
11.Rc1 Nb4 12.Qd2 Bd7 13.Nb3 d5 14.e5 Ne4 15.Nxe4 dxe4 16.Qd4
  ²  (0.42)   Depth: 11/32   00:03:38  70138kN

========

    depth    time (seconds)

      7                  1
      8                  5
      9                  9
     10                 37
     11                218

Let's assume an effective branching factor of 3, and continue guessing the
times:

     12                654
     13               1962
     14               5886
     15              17658
     16              52974
     17             158922
     18             476766
     19            1430298
     20            4290894

4290894 seconds on a 733 Mhz processor.

Now, processor speedup is linear, i.e., a 2x faster processor will result in the
time being reduced by half. Also tournament time control is about 3 minutes per
move (180 seconds). So:

processor_you_need = 733 * (4290894 / 180) = 17473473.9 MHz

That is, for reaching 20 plies on the above position in a tournament time
control using Crafty, you will need a processor of something like 17000 GHz
(about 16.6 THz). And today we "only" have 3 GHz!

That's why I believe more in algorithmic speedup (exponential) rather than
hardware speedup (linear).



This page took 0 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.