Author: Vincent Diepeveen
Date: 14:45:10 04/13/01
Hello, gcc in past always was very slow for me. 2.95.2 especially was at a PIII-cumine core around 10.8% slower. Note Xeons seem not made from the same core so always were slower as celeron and PIII for me. PIII itself nearly 20% faster as PII katmai core. Xeon 7% faster as PII and celeron 7.5% faster. However that's all single cpu measured. When running dual then very interesting is the size of the program in the L1,L2 caches (when talking about DIEP that is). Single cpu gcc 2.95.3 is only 8% slower as visual c++ 6.0 sp5 now. Using compile options: -O5 -march=pentiumpro -mcpu=pentiumpro -fno-gcse -fforce-mem note -mcpu=pentiumpro might be not needed as march might already mean that. Now if i take the exact same executable and compare it with 2.95.2, then the executable is 100kb smaller!! I was very happy about it but it started to amaze me when i started running it. Of course when running parallel it's very hard to measure very exactly differences. So i did some quite long testruns and compared at the same depth the nodes a second that visual and gcc got. gcc 2.95.3 : 68311 msvc 6.0 sp5 : 70110 Now THAT is amazing as it is even from head only about 3% difference in speed. Hardware used was my dual 800PIII. 256mb ram and hashsize i chose was 150mb. In the test no EGTBs were loaded. So GCC is getting better again! HIEP HIEP HORAY! Best regards, Vincent
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.