****************************
M. Ray McFerron
302 E. 26th Ave
North Kansas City, MO 64116
**   http://cctr.umkc.edu/~mmcferron  **
****************************

----------
Actually this is incorrect.  Although I don't have the f/p benchmark results
directly in front of me, the latest version of macworld (I think the May
1998) shows a G3/300 greatly outperforming a 9600/300.....

M. Ray McFerron


>From: Richard Dobson 
>To: Piche Jean 
>Cc: csound@maths.ex.ac.uk
>Subject: Re: csound realtime
>Date: Wed, Apr 8, 1998, 3:18 PM
>

>I recall reading in one of the computer magazines that the fpu performance
of the
>G3 processor was in fact somewhat less than that of the 'normal' PowerPC.
Most
>journals pay little attention to this sort of detail (and manufacturers
don't
>exactly encourage them to!), but it is obviously important when running
Csound;
>so it would be useful to have some proper f/p benchmarks. The G3 sounds
like good
>value (relatively...), but if it is sluggish on f/p it is rather less
attractive!
>
>Richard Dobson

On Wed, 8 Apr 1998, M. Ray McFerron wrote:

> Actually this is incorrect.  Although I don't have the f/p benchmark results
> directly in front of me, the latest version of macworld (I think the May
> 1998) shows a G3/300 greatly outperforming a 9600/300.....

 here's what i got (note that the 604e is clocked 350MHz) as far as the
chips goes: 

 	PPC750@300 (aka G3): SPECint95 13.2 SPECfpu95 8.5
 	PPC604e@350        : SPECint95 14.6 SPECfpu95 9.0

 looks like quite similar performance; similar int/float ratio. But these
"raw" numbers don't say it all. There's cache size & ratio, bus speed,
etc. (The G3 upgrade cards vary a *lot* in price depending on the cache
you get. ). The Macworld "real world" benchmark suite gives the G3/750@233
a 4.5 and the 9600/604e@233 a 3.3.  (Of course the G3 is way cheaper than
the 9600). That benchmark is a mixture of CPU, FPU, disk and video
performance. 

 but these ways of measuring speed do not necessarily apply to csound. the
only way to know about how csound will perform is to try it, but the
csound benchmark is a little on the old side and compiles so fast that it
is meaningless. If someone has given some thought to a "modern" suite
testing different aspects of csound (raw calculation, disk access,
realtime limits) and has an orchestra handy, i'll be happy to subject a
G3@233, a 604e@180 and a 586@166 (linux) to it; maybe some SGI's R10000 and
R5000 if there's interest (and while we're at it, a SHARC...). I could
also play with the cache size to see how it's relevant to csound
calculations. But i beleive some time has to be spent designing these
tests (by someone who knows how csound really operates) so they really are
meaningfull. Or we could just compile some long synthesis pieces and
compare the results? I guess different configs should also be tested to 
se if there is room for optimisation (i know the motorola math 
library gave a nice boost to MacOS7.5-6).

 however, this will not settle the neverending OS issue, although a 
comparison of mklinux and macos would be interesting (as well as a 
linux/windows one) so we could see how the OSes affect performance 
(multitasking & stability aside).

 anyone interested in a "serious" csound benchmarking effort?

						ALex Burton.


> >From: Richard Dobson 
> >To: Piche Jean 
> >Cc: csound@maths.ex.ac.uk
> >Subject: Re: csound realtime
> >Date: Wed, Apr 8, 1998, 3:18 PM
> >
> 
> >I recall reading in one of the computer magazines that the fpu performance
> of the
> >G3 processor was in fact somewhat less than that of the 'normal' PowerPC.
> Most
> >journals pay little attention to this sort of detail (and manufacturers
> don't
> >exactly encourage them to!), but it is obviously important when running
> Csound;
> >so it would be useful to have some proper f/p benchmarks. The G3 sounds
> like good
> >value (relatively...), but if it is sluggish on f/p it is rather less
> attractive!
> >
> >Richard Dobson
> 



                                                                   Alex.

   
                                          "the map is not the territory"  
                                                             (korzybski)