Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: crafty on Athlon 64

Author: Vincent Diepeveen

Date: 20:25:26 12/31/02

Go up one level in this thread


On December 30, 2002 at 23:35:05, Matt Taylor wrote:

There is a very simple thing in processorland. Test before
making claims. Never buy shares because of drumming of a manufacturer.

Only do so after testing speed of it :)

>On December 30, 2002 at 20:19:32, Vincent Diepeveen wrote:
>
>>On December 30, 2002 at 20:15:01, Matt Taylor wrote:
>>
>>I was under impression this was a specint2000 test at the
>>specint homepage.
>>
>>Still i find personally 868 a very IMPRESSIVE score for a 1.2Ghz
>>thing.
>>
>>Compare to 2.2Ghz P4. having LESS than that.
>>
>>Note that 64 bits code is slower on the AMD processor than 32 bits
>>code. I'm not sure you want to use the 64 bits code for 99% of the
>>crafty code.
>
>Uh, what? Generally speaking, 64-bit code runs at the same speed as 32-bit code
>on the Hammer. It runs with a few notable exceptions at the same speed as on its
>little brother Athlon. Please read the Page 19 of the K8 optimization
>presentation by Tim Wilkens before making such claims.
>
>http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Optimization_-_Tim_Wilkens.pdf
>
>I will clue you in that a lot of computation is focused on the ALU, and all the
>ALU operations run at 1 clock latency (with maximum throughput of 3 ops/cycle).
>
>Also, the location of the benchmarks is irrelevant. SPEC did not run the
>benchmark for Hammer. I could not find K8 results on their webpages.
>
>>>On December 30, 2002 at 19:19:00, Vincent Diepeveen wrote:
>>>
>>>>On December 30, 2002 at 19:13:36, Bob Durrett wrote:
>>>>
>>>>>On December 30, 2002 at 12:23:59, Vincent Diepeveen wrote:
>>>>>
>>>>>>On December 30, 2002 at 11:55:51, Sean Mintz wrote:
>>>>>>
>>>>>>>I'm interested to know if they compiled crafty for 64 bits. I doubt that because
>>>>>>>the 32 bit athlons running at the same speed get around 650. That's a 25%
>>>>>>>increase. Isn't this the improvement they said we would get from running 32 bit
>>>>>>>applications on the athlon 64? Anyone else have any ideas?
>>>>>>
>>>>>>Of course it was a 64 bits compile. They are not idiots. they have a multi
>>>>>>billion dollar reason to compile it for 64 bits.
>>>>>>
>>>>>>Rethink your statement!
>>>>>>
>>>>>>Suppose someone from the testdepartment then tells his boss that the
>>>>>>billions of investments into their 64 bits cpu were shown a bit worse
>>>>>>because they only ran software in 64 bits.
>>>>>>
>>>>>>How many seconds before entire testdepartment is fired?
>>>>>>
>>>>>>0.01 if i was that manager.
>>>>>>
>>>>>>Of course the testdepartment optimized the executable as far as they could
>>>>>>for the 64 bits processor. Anything that was possible until now has been done
>>>>>>let me assure you that.
>>>>>>
>>>>>>The fastest executables with production compilers can be found at specint
>>>>>>of course. Only beta versions of copmilers which didn't make it into the
>>>>>>mainstream are possibly not allowed to get used for specint (which is
>>>>>>a good thing).
>>>>>>
>>>>>>So if there is anything out there yet that allows to compile crafty to
>>>>>>64 bits somehow, be sure they used it.
>>>>>>
>>>>>>Best regards,
>>>>>>Vincent
>>>>>
>>>>>Vincent, please forgive me for saying this, but:
>>>>>
>>>>>You have not presented one shread of evidence to prove that they used 64 bit.
>>>>>All you have presented is sheer speculation!
>>>>>
>>>>>What are the FACTS?  Do you have any?
>>>>>
>>>>>Sorry about that, but someone had to say it.
>>>>>
>>>>>Bob D.
>>>>
>>>>As you can see in specint the different companies which earn billions with
>>>>processors will of course let their processor look like the best. They optimally
>>>>prepare the software for their processor like they want to, of course under the
>>>>rules there are.
>>>>
>>>>That means in short that it is more than naive to suppose they are not using 64
>>>>bits when it can speed them up.
>>>>
>>>>There is billions at stake here.
>>>>
>>>>Please do not tell me the companies are idiots in this sense.
>>>>
>>>>Best regards,
>>>>Vincent
>>>
>>>Yes, but I didn't see any evidence that AMD ran the benchmark either. From what
>>>I could tell, it seemed like a 3rd party got a Clawhammer sample and ran the
>>>benchmark.
>>>
>>>I'm sure AMD wants to look good -- but do other people care as much about AMD's
>>>reputation?
>>>
>>>-Matt



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.