Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty19.15 Benchmarks on Anandtech Xeon 3.6 vs Opteron 150

Author: Aaron Gordon

Date: 18:37:01 08/13/04

Go up one level in this thread


On August 13, 2004 at 14:09:35, Anthony Cozzie wrote:

>On August 13, 2004 at 12:44:21, Aaron Gordon wrote:
>
>>On August 13, 2004 at 11:55:12, Anthony Cozzie wrote:
>>
>>>On August 13, 2004 at 02:12:04, Aaron Gordon wrote:
>>>
>>>>All the bad press finally got him redo the review. I told'em a few cflags he
>>>>could use to make performance better.. unfortunately he didn't do the PGO
>>>>optimization. I think he thought because I said I used GCC 3.4.1 that it was
>>>>only available in 3.4.1, who knows..
>>>>
>>>>Anyway, looks like the 64bit stuff on the Nacona offers no performance benefit
>>>>(only addressing). Would fit with the rumors I heard a few months back. We'll
>>>>see I guess. Supposedly they just "got it working" for now and will fix the
>>>>performance issue later.
>>>>
>>>>Anyone know any hard facts about the 64bit Xeon stuff? Broken?
>>>
>>>I thought Nocona was an implementation of of AMD64 ?
>>>
>>>anthony
>>
>>From what I hear it is, however supposedly it only has the addressing of a 64bit
>>chip (ie. can use that much ram, can use 64bit integers/fp/etc), but cannot do
>>them as fast as it is suppose to be. The way I heard it described was like it
>>was using the 32bit registers to do all of the stuff.. similar to a regular P4
>>trying to do a __int64, long long, or double.
>>
>>I can't verify this, but from the results that Anandtech did it sure looks like
>>the rumor may be true. The other part of the rumor was that Intel was going to
>>fix it soon. I think this was a quick add-on after they saw x86-64 was the next
>>big thing and didn't have time to rework the core to get it functioning 100%.
>>
>>I guess we'll see in the months ahead...
>
>Hmm, that would definitely be possible.  And _something_ is wrong with the
>Crafty benchmark because in 32 bit mode the opteron is not _that_ much faster
>than the P4 for Crafty.
>
>anthony

The test was compiled in 64bit mode, which is why the Opteron is getting over 2
million nodes/second. The reason I say the Xeons 64bit stuff could be broken is
it was compiled the same way and gains nothing apparently.



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.