Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question: Itanium Info

Author: Eugene Nalimov

Date: 20:41:21 12/11/03

Go up one level in this thread


On December 11, 2003 at 17:14:26, Brian Richardson wrote:

>On December 11, 2003 at 15:36:49, Eugene Nalimov wrote:
>
>>On December 11, 2003 at 13:44:59, Brian Richardson wrote:
>>
>>>On December 10, 2003 at 20:25:43, Eugene Nalimov wrote:
>>>
>>>>On December 10, 2003 at 19:24:12, Brian Richardson wrote:
>>>>
>>>>>On December 10, 2003 at 18:38:03, Eugene Nalimov wrote:
>>>>>
>>>>>>On December 09, 2003 at 20:22:16, Robert Hyatt wrote:
>>>>>>
>>>>>>>On December 09, 2003 at 16:12:46, Brian Richardson wrote:
>>>>>>>
>>>>>>>>On December 09, 2003 at 09:52:34, Robert Hyatt wrote:
>>>>>>>>
>>>>>>>>>On December 08, 2003 at 20:59:26, Slater Wold wrote:
>>>>>>>>>
>>>>>>>>
>>>>>>>>snipped
>>>>>>>>
>>>>>>>>>>Ok, that's the itanium doing 32.  Anyone got anything with it doing 64?  Or did
>>>>>>>>>>it suck there too?
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>The original was not very good.  Itanium-2 (Mckinley) is _very_ good.  Close
>>>>>>>>>to the opteron even though it is clocked at 1/2 the opteron's speed.
>>>>>>>>
>>>>>>>>Actually, McKinley was also pretty poor, IIRC.  I had emailed Bob some Crafty
>>>>>>>>bench command test results.  Now the 3rd generation Madison is much better.
>>>>>>>
>>>>>>>
>>>>>>>Eugene was close to 1M nodes per second at 1ghz.  I don't think I have his
>>>>>>>numbers immediately handy but he might supply them again...
>>>>>>
>>>>>>I don't remember exact numbers, but on 1GHz Itanium2 (McKinley) Crafty got
>>>>>>something like 900-1000knps when executing "bench" command. Not great, but
>>>>>>reasonable good number.
>>>>>>
>>>>>>On 1.5GHz Itanium2 (Madison) Crafty is getting 1,357knps.
>>>>>>
>>>>>>If necessary I can send executable to Bob, so any volunteer can run his/her own
>>>>>>tests.
>>>>>>
>>>>>>Thanks,
>>>>>>Eugene
>>>>>
>>>>>The 900Knps was for 2 CPUs; 1 CPU was about 500Knps, according to the log files
>>>>>(note for Crafty 18.15, Intel compiler, no assembler, no profiling).
>>>>>
>>>>>Non-recompiled 32bit binary was _much_ slower, of course.
>>>>
>>>>Ok, I found 900MHz/1.5Mb cache system nearby. Here are the results:
>>>>
>>>>D:\Documents and Settings\eugenen>\\eugenen6\crafty\wcrafty.exe
>>>>
>>>>Initializing multiple threads.
>>>>System is SMP, not NUMA.
>>>>EPD Kit revision date: 1996.04.21
>>>>unable to open book file [./book.bin].
>>>>book is disabled
>>>>unable to open book file [./books.bin].
>>>>
>>>>Crafty v19.6 (1 cpus)
>>>>
>>>>White(1): bench
>>>>Running benchmark. . .
>>>>......
>>>>Total nodes: 100409437
>>>>Raw nodes per second: 749324
>>>>Total elapsed time: 134
>>>>SMP time-to-ply measurement: 4.776119
>>>>White(1): quit
>>>>
>>>>I expect 1GHz/3Mb cache system to be ~20% faster -- 10% due to higher frequency,
>>>>and 10% due to larger cache (or higher cache associativity -- I reported effect
>>>>of 1.5Mb cache vs. 3Mb cache here some time ago). 750knps*1.2 == 900knps, so it
>>>>will be roughly the number I gave from memory...
>>>>
>>>>Thanks,
>>>>Eugene
>>>
>>>And the version 18.15 results (19.6 was not out back in January)?
>>
>>I don't have old Crafty sources nearby, and cannot connect to UAB site (some
>>configuration problems on MS campus). Bob already wrote that there should be no
>>difference in nps, and 750knps on 900MHz/1.5Mb system agrees with 900-1000knps
>>on 1GHz/3Mb system I gave from memory.
>>
>>It looks that you used inferior compiler...
>>
>>Thanks,
>>Eugene
>
>You (and Bob) are correct that there is not much of a nps difference
>between versions 18.15 and 19.05(06).  I was thinking of other version
>differences that were larger.
>
>In any case, might you be running with a "superior compiler" that the rest
>of us don't have ready access to :)

I am pretty sure that publicly available to all MSDN subscribers (as part of
PSDK) 64-bit Microsoft compiler will generate decent code for Crafty. Maybe
10-12% slower than the latest one. That still should put Crafty on 1GHz Itanium2
into 800+ knps range when compiled with POGO.

Thanks,
Eugene



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.