Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: To SMP or not to SMP what's the answer?

Author: Charles Worthington

Date: 07:55:16 04/13/03

Go up one level in this thread


On April 13, 2003 at 08:09:03, Vincent Diepeveen wrote:

>On April 12, 2003 at 23:04:17, Charles Worthington wrote:
>
>>On April 11, 2003 at 21:39:55, Vincent Diepeveen wrote:
>>
>>>On April 11, 2003 at 18:07:30, Charles Worthington wrote:
>>>
>>>>On April 11, 2003 at 06:37:41, Vincent Diepeveen wrote:
>>>>
>>>>>On April 11, 2003 at 00:31:52, Charles Worthington wrote:
>>>>>
>>>>>>On April 11, 2003 at 00:09:23, Jay Urbanski wrote:
>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>It's simply wrong, or you have deliberately chosen to ignore all consumer
>>>>>>>>electronic devices and most PCs.
>>>>>>>>
>>>>>>>>I don't even know if the vast majority of processor produced in the world are 16
>>>>>>>>bits or 32 bits ones. Maybe the majority is 8 bits processors.
>>>>>>>>
>>>>>>>
>>>>>>>Perhaps if you include all embedded processors in the world.. but even that's
>>>>>>>doubtful.  But on the other hand I doubt your car or your refrigerator are going
>>>>>>>to be running a chess engine.
>>>>>>>
>>>>>>>SMP / SMT processing is definitely on the rise and you *will* see it become much
>>>>>>>more common on both the desktop and general purpose server machines.  IBM's
>>>>>>>POWER4 is already SMP on a chip, Itanium will be soon, as will Opteron in a few
>>>>>>>years.  Intel will follow suit on the desktop as they have with Hyperthreading
>>>>>>>already.  It's a cheap way to get more processing power out of the available
>>>>>>>silicon - so why not?
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>Not to mention the fact that they will soon reach a ceiling on how much speed
>>>>>>they can get out of a single cpu and once that happens multiprocessing will no
>>>>>>longer be a luxury...it will be a necessity.
>>>>>
>>>>>I disagree here. So far there is no indication they will reach a ceiling.
>>>>>
>>>>>SMT/HT is a big sales argument and it is possible to make it now because we are
>>>>>nowadays at 0.13 micron. In 0.18 or .35 micron this would have been harder to do
>>>>>against the same price.
>>>>
>>>>
>>>>
>>>>Actually there is no choice but for them to reach a ceiling eventually...They
>>>
>>>that is a very cheap statement to do actually.
>>>
>>>So far they managed to get faster each 2 years about 2 times. Overall seen even
>>>more than that.
>>>
>>>So many things can get improved IPC can get improved. caches. Prediction
>>>algorithms. All very hard to improve but they will improve all bit by bit.
>>>
>>>when hardware gets very tiny then i love to see 32 cpu's at a single
>>>processor-die :)
>>>
>>>>are limited by the size of the atom and other factors such as conductivity which
>>>>are controlled by the laws of physics. No matter how smart the engineers are the
>>>>current technology does in fact have limits....They are the_laws_of
>>>>physics...not the_theory_of physics.
>>>>
>>>>respectfully, Charles
>>
>>
>>Yes i agree that there is still significant room for improvement. However, it is
>>still a question of which will come first....The technology to improve single
>>cpu's or the need to move on to multiprocessing in order to get the performance
>>increases that the public have become accustomed to. If multiprocessing winds up
>>being the solution then the cost of dual cpu systems will drop dramatically as
>>they become commonplace. Right now it is looking like a toss up as to which will
>>come first. Personally I am hoping to see the improvements in the single cpu's
>>because then that will make them all the more deadly in the multiprocessor
>>systems. Eventually, however, there is still a ceiling that will have to one day
>>be dealt with by multiprocessing or a completely new technology which does not
>>rely on the transistor.
>
>Let's skip that last remark for now.
>
>It is trivial that a cpu in itself is already a parallel thing. Already for many
>years instructions get executed simultaneously.


I am no cpu expert Vincent, but I was under the impression that a single thread
switched back and forth between processes at a speed which only makes
it_appear_to be performing simultaneous tasks. A thread can only run_one_program
at a time based on what I have read.

Respectfully,
Charles



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.