Author: Moritz Berger
Date: 14:16:50 01/03/98
Go up one level in this thread
On January 03, 1998 at 01:00:07, Albert Silver wrote: >I tested Rebel 9, Fritz 5.01, and Hiarcs 6 (Fritz module) with the >Louguet Test II on my K6/233 (with 512k internal pipeline cache). Fritz >5 and Hiarcs 6 were tested with 12 Megs of hash in Win95 (please don't >tell me that Fritz 5 needs 100 Megs to run ideally as I still think >that's the stupidest thing I ever heard of) see figures below and take into account that more hash tables doesn't only mean faster but more accurate search. with Fritz's automatic test >processing with a setting of 2 plies (meaning it tests for an extra ply >to verify the program doesn't change it's mind) I used the same settings. <snip> Here's my results, I tested Fritz both with reasonable (i.e. 98304KB) and "unreasonable" :-))) (12288KB) hash tables. Hiarcs seems mostly insensitive to hash tables beyond 12MB (compare solution times on AMD K6 with 12MB vs. P5MMX with 98 MB - no noteable performance improvement for the P5 on longer pondering times despite bigger hash tables). Prog Fritz Fritz Fritz Junior Hiarcs Hiarcs Ver/cpu 5.01 5.01 5.01 4.6beta 6 Eng. 6 Eng. P233MMX P233MMX K6/233 P233MMX P233MMX K6/233 Hash 98304KB 12288KB 12288KB 98304KB 98304KB 12288KB POS 01 d6 12 12 17 13 3 3 POS 02 ...Fb4 1 1 2 2 1 1 POS 03 Dc5 POS 04 ...e5 3 2 2 POS 05 ...Fb5 1 1 128 31 28 POS 06 ...e5 28 32 19 154 18 19 POS 07 Cd1 4 6 6 POS 08 Fh3 195 247 289 1 1 POS 09 Dd4 11 10 11 44 44 39 POS 10 ...Ff8 18 POS 11 h4 392 74 71 POS 12 ...Tb6 5 5 5 418 414 POS 13 ...Cxa2 3 3 10 POS 14 d5 467 466 CMB 01 Cxd6 2 2 4 4 1 1 CMB 02 Dxh7 1 1 2 85 1 1 CMB 03 ...Df3 2 2 2 8 1 1 CMB 04 exf6 4 4 5 3 1 1 CMB 05 Txh7 10 10 8 32 4 4 CMB 06 exf6 32 36 41 4 4 5 CMB 07 ...Txc3 6 6 7 25 15 16 CMB 08 Cf6 11 11 14 24 83 75 CMB 09 ...Txd2 46 51 65 44 2 2 CMB 10 Fxh6 240 410 349 CMB 11 Cxh7 CMB 12 e5 FIN 01 f6 2 2 3 6 7 7 FIN 02 f5 1 1 1 3 1 1 FIN 03 ...Fxe4 14 14 38 100 61 56 FIN 04 ...h3 1 1 68 3 24 FIN 05 a6 206 278 335 70 18 17 FIN 06 ...f4 57 33 34 FIN 07 Fb4 92 125 194 162 123 120 FIN 08 c5 FIN 09 Rg4 Pts POS 49% 48% 44% 43% 58% 60% Pts CMB 67% 67% 68% 67% 72% 74% Pts FIN 52% 52% 44% 59% 61% 52% ELO 2485 2480 2450 2480 2570 2555 Time used 686 855 1267 1501 1854 1740 Ratio K6/P5 1.85 1.48 0.95 Summary: The Pentium MMX is only roughly 95% as fast for Hiarcs 6 on this test suite than the K6 at the same clock speed. Fritz runs 1.5 times faster on the MMX, additional hash tables give it another boost (almost as fast as a hypothetical AMD K6 400 :-)). Moritz P.S.: Hiarcs 6 finds FIN 04 on my machine both with 12288 and 98304 KB hash tables. Albert, could you verify your result for FIN 04 ???
This page took 0.03 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.