|
From: | Marco Scheurer |
Subject: | Re: [gnugo-devel] GNU Go 3.3.23 |
Date: | Wed, 16 Jul 2003 21:10:27 +0200 |
On Wednesday, July 16, 2003, at 07:54 PM, <address@hidden> wrote:
On my Mac, "time make regression" gives the following results: 3.3.23 1050.970u 5.850s 23:24.91 75.2% 0+0k 1+23io 0pf+0w 3.2 1249.210u 6.040s 24:16.30 86.1% 0+0k 0+23io 0pf+0w All this using the defaults compile and run options. However, I'm not sure it proves anything since the regression tests themselves have changed. Is there a constant benchmarking test?Are you are using the 3.2 testsuite with 3.2 and the 3.3.23 testsuite with 3.3.23? If so I am surprised that 3.3.23 shows less user time than 3.2. The reason this is surprising is that there are MANY more tests in 3.3.23 than in 3.2. The long fifth batch consists of new tests. The first four batches are mainly old tests. I suppose I should go back and run the 3.2 regressions and see how long they take.
More results: On 3.2 first_batch 3.3.23 1025.510u 7.310s 40:04.13 42.9% 0+0k 0+16io 0pf+0w 3.2 1249.210u 6.040s 24:16.30 86.1% 0+0k 0+23io 0pf+0w On 3.3.23 first_batch 3.3.23 1050.970u 5.850s 23:24.91 75.2% 0+0k 1+23io 0pf+0w3.2 N/A (3.2 does not know about the reg_genmove command which is used in some test files)
Of course I understand that timing these tests is not necessarily a good measure of the time spent in a real game, but so far, on my system, it seems that 3.3.23 is faster. I haven't used any option, so I assume that both 3.2 and 3.3.23 were running at level 10.
I'm now doing the same on 3.2 all_batches. marco
[Prev in Thread] | Current Thread | [Next in Thread] |