Author: Keith Evans
Date: 21:12:33 07/31/03
Go up one level in this thread
On July 31, 2003 at 23:14:28, Slater Wold wrote:
>On July 31, 2003 at 21:34:13, Dann Corbit wrote:
>
>>On July 31, 2003 at 19:41:38, Ed Schröder wrote:
>>
>>>On July 31, 2003 at 19:29:14, Slater Wold wrote:
>>>
>>>>On July 31, 2003 at 18:41:50, Dann Corbit wrote:
>>>>
>>>>>On July 31, 2003 at 08:27:55, Slater Wold wrote:
>>>>>
>>>>>>On July 31, 2003 at 02:45:21, Ed Schröder wrote:
>>>>>>
>>>>>>>On July 30, 2003 at 18:50:33, Slater Wold wrote:
>>>>>>>
>>>>>>>>Time Control = 5/5
>>>>>>>>Shreder 7.04 - Dual P4 3.06Ghz - +3 -15 =1 (10 loses on time)
>>>>>>>>Deep Fritz 7 - Dual AMD 2.3Ghz - +1 -8 =3 (11 loses on time)
>>>>>>>>Time Control = 15/5
>>>>>>>>Deep Fritz 7 - Dual AMD 2.3Ghz - +0 -0 =1
>>>>>>>
>>>>>>>
>>>>>>>Slate,
>>>>>>>
>>>>>>>You have been playing this under Chesspartner I assume?
>>>>>>>
>>>>>>>I have no idea, everything looks fine here but I haven't played Rebel on ICC
>>>>>>>yet. Maybe it is related to the permanent brain, that wouldn't be the first time
>>>>>>>and you should turn off pondering for this version playing ICC.
>>>>>>>
>>>>>>>
>>>>>>>>Total of 12 hours on ICC = +4 -23 =5 (And 23 loses on time.)
>>>>>>>
>>>>>>>That must have been frustrating for you :)
>>>>>>>
>>>>>>>Anyway, thanks for pointing this out.
>>>>>>>
>>>>>>>Ed
>>>>>>>
>>>>>>>
>>>>>>>>Games will follow.
>>>>>>
>>>>>>Using 200MB hash, I haven't lost on time in 15 games.
>>>>>>
>>>>>>I think using 400MB hash is what was locking up the search.
>>>>>>
>>>>>>I have 1GB of RAM, but....it does seem that the hash was the problem.
>>>>>>
>>>>>>And now it's not losing on time, and doing much better against it's opponents.
>>>>>>:)
>>>>>
>>>>>400 MB hash in a blitz game!
>>>>>
>>>>>That is always going to be a problem. Even if it could move on time, it's going
>>>>>to play crappy.
>>>>>
>>>>>Think of how long it takes to clear out 400 MB of memory.
>>>>
>>>>On a P4? Not long.
>>
>>Longer than you think, I'll guess.
>>Try this program:
>>
>>#include <assert.h>
>>#include <stdio.h>
>>#include <string.h>
>>#include <stdlib.h>
>>#include <time.h>
>>
>>
>>int main(void)
>>{
>> clock_t start,
>> end;
>> size_t i;
>> size_t ramsize = 400U * 1024U * 1024U;
>> void *hash = malloc(ramsize);
>> if (hash) {
>> assert(ramsize > 4e8); /* some compilers might overflow */
>> start = clock();
>> for (i = 0; i < 100; i++)
>> memset(hash, 0, ramsize);
>> end = clock();
>> printf("Each clearing of the hash took %.0f seconds\n",
>> (end - start) * 1.0 / CLOCKS_PER_SEC);
>> }
>> return 0;
>>}
>>
>>
>>>>But Rebel uses a swap file, which is probably a bit of a problem. Even on an
>>>>U320 SCSI drive.
>>>
>>>During blitz games Rebel only uses a small part of the hash table, no swapping
>>>possible. See the PARAM.TXT file.
>>
>>But apparently, it's irrelevant.
>
>
>I stand corrected. ;)
>
>49 seconds.
What am I missing? It seems like this should be 49 / 100 seconds...?
There's just no way that 49 seconds makes any sense. That would be a memory
bandwidth of 400/49 = 8 MB/s which is obviously wrong.
-Keith
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.