Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Pentium 4 and hyperthreading (to Charles Worthington)

Author: Robert Hyatt

Date: 15:04:19 02/19/03

Go up one level in this thread


On February 19, 2003 at 16:41:09, Steffen Basting wrote:

>On February 19, 2003 at 16:18:54, Charles Worthington wrote:
>
>>On February 19, 2003 at 16:13:58, Steffen Basting wrote:
>>
>>>Hi!
>>>No, I would say Martin is right. You can do that with rather practical numbers:
>>>
>>>One processor, ht disabled: 1.000 nps.
>>>=> Two processors, ht disabled: 2.000 nps.
>>>
>>>One processor, ht enabled: 1160 nps (ht disabled + 16%).
>>>=> Two processors, ht enabled: 2.320 nps.
>>>
>>>and 2.000 * 1.16 = 2.320, so your speed up is 16% for both cpus.
>>>
>>>
>>>Regards, Steffen
>>
>>
>>actally the speedup is doubled when you are dealing with dual processors. You
>>are using reverse mathematics to arrive at an incorrect answer. I honestly think
>>Dr. Hyatt knows his math.
>
>There's no doubt that Dr. Hyatt knows his math - I just cannot see the "bug" in
>my example. If the speed-up increases in the way you describe (n proc * 16%), it
>would mean that with 16 processors ht enabled you arrive at 256%. So you are
>faster than > 32 processors with ht disabled. This doesn't seem to be correct...
>
>Regards, Steffen


this is semantics.  It depends on how you define "percent faster".

IE if you take a one processor machine as the benchmark, SMT will make it 20-30%
faster with Crafty.  If you take the two processor machine as the benchmark, it
will run
20-30% faster with SMT.  But that is 20-30% faster than two processors without
SMT.

That is the only way I really calculate this...  What does the extra SMT thread
gain, and the
answer is adding that extra thread makes that physical processor 20-30% faster
than it
was without it.



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.