portable computation time estimation?

portable computation time estimation?

Post by NortonN » Fri, 27 Apr 2001 17:42:23



hi all,

       I am a problem in writing portable code about computation time
estaimation at various CPU speed.
      If a job can done in 2 second on PIII 733Mhz cpu
Can i estimate correctly how long it will take on PIII 900Ghz cpu??

Regards,
jkng

 
 
 

portable computation time estimation?

Post by Juergen Hannappe » Fri, 27 Apr 2001 18:41:59



> hi all,

>        I am a problem in writing portable code about computation time
> estaimation at various CPU speed.
>       If a job can done in 2 second on PIII 733Mhz cpu
> Can i estimate correctly how long it will take on PIII 900Ghz cpu??

                                                            ^^^
If you talk about the factor 900/733 = 1.227 in speed the estimation
1.6288 seconds should be fairly accurate. If you talk about a factor
1277 which you would get with a 900 Giga-Hertz CPU i would assume that
it would take longer than 1.6 milliseconds, but probably there will
never be a 900GHz PIII.

Greetings,
        Juergen  

--
Dr. Juergen Hannappel  Office: W148 Phone: +49 228 73 2447  FAX +49 228 73 7869

http://lisa2.physik.uni-bonn.de/~hannappe   Nussallee 12, D-53115 Bonn, Germany
CERN: Phone: +412276 76461 Fax: 77930 Bat. 892-R-A13 F-01631 CERN CEDEX, France

 
 
 

portable computation time estimation?

Post by Josef Moeller » Fri, 27 Apr 2001 23:19:50




> > hi all,

> >        I am a problem in writing portable code about computation time
> > estaimation at various CPU speed.
> >       If a job can done in 2 second on PIII 733Mhz cpu
> > Can i estimate correctly how long it will take on PIII 900Ghz cpu??
>                                                             ^^^
> If you talk about the factor 900/733 = 1.227 in speed the estimation
> 1.6288 seconds should be fairly accurate. If you talk about a factor

I doubt that this simple calculation holds for more than just a small
handfull of processor-pairs. There is more to processor speed than just
clock speed. A certain processor core may need more cycles for a given
calculation than another processor core.
Also the speed of memory accesses cannot simply be ignored: a faster
processor may possibly have to wait (conceptually) longer for data from
memory than a slower processor.

Quote:> 1277 which you would get with a 900 Giga-Hertz CPU i would assume that
> it would take longer than 1.6 milliseconds, but probably there will
> never be a 900GHz PIII.

Hmm, you sound like Scott Nudds ... at least you write "probably"! SCNR

--
Josef M?llers (Pinguinpfleger bei FSC)
        If failure had no penalty success would not be a prize
                                                -- T.  Pratchett

 
 
 

portable computation time estimation?

Post by Tim Robert » Sat, 28 Apr 2001 15:27:21



>                                                            ^^^
>If you talk about the factor 900/733 = 1.227 in speed the estimation
>1.6288 seconds should be fairly accurate. If you talk about a factor
>1277 which you would get with a 900 Giga-Hertz CPU i would assume that
>it would take longer than 1.6 milliseconds, but probably there will
>never be a 900GHz PIII.

If Moore's law continues to hold, just wait 15 years.

At 900GHz, electrons will take three clock cycles to travel a millimeter.
Two ends of a centimeter-wide chip will be 30 clock cycles off!  That'll
make circuit design a *...
--

  Providenza & Boekelheide, Inc.

 
 
 

portable computation time estimation?

Post by Tom Robert » Wed, 02 May 2001 04:03:24



>       If a job can done in 2 second on PIII 733Mhz cpu
> Can i estimate correctly how long it will take on PIII 900Ghz cpu??

        [I ignore the typo, Ghz => Mhz]

It is almost hopeless to obtain accurate estimates without doing a careful
modeling of the program and the hardware. For some programs the best estimate
would be 2 seconds, and not 733/900*2 seconds. For other programs the latter
would be better. For some hardware, the 900 MHz CPU might actually take
LONGER than 2 seconds (because of some other bottleneck not present in the
700 MHz hardware).

For example, time memcpy(). For small data blocks it will track accurately
the CPU clock speed (because all memory remains in cache during the timing).
For very large data blocks it will track _MEMORY_ clock speed (which is
presumably the same for the two CPUs), because the cache is thrashed by the
copy. Intermediate-sized blocks will be somewhere in between. This ignores the
fact that simple timing may not really time the entire operation (i.e. it may
only time the cache->cache copy, or at most the memory(->cache)->cache copy,
and not the later cache cast-outs to physical memory).

Trying to compare a PIII to a PIV or an Athlon is difficult (at best). Trying
to compare them to a PowerPC is hopeless (for a given clock speed, a PowerPC
6400 can be vastly faster than a PIII...).


 
 
 

portable computation time estimation?

Post by Rolf Magnu » Thu, 03 May 2001 23:30:50




>>                                                            ^^^
>>If you talk about the factor 900/733 = 1.227 in speed the estimation
>>1.6288 seconds should be fairly accurate. If you talk about a factor
>>1277 which you would get with a 900 Giga-Hertz CPU i would assume that
>>it would take longer than 1.6 milliseconds, but probably there will
>>never be a 900GHz PIII.

> If Moore's law continues to hold, just wait 15 years.

Who wants a PIII CPU in 15 years? It's only 32bit.