At:

http://www.iro.umontreal.ca/~pigeon/pub/PovPage/PovPage.html

Include:

-- Optimized for Pentium 2 processor and Windows NT (works on 95/98,
though)
-- ETA on status bar
-- float pps, i.e., you can see 0.03 pps instead of 0 PPS.  (someone
asked for it... but... ?)
-- pixel in row progress status
-- minor bug fixes (palette and refresh of rendering window, default ini
file loading,
    etc.)

Best,

    S.

----------------------------------------------------------
Steven Pigeon                     Ph. D. Student.
University of Montreal.
pig### [at] iroumontrealca           Topics: data compression,
pig### [at] jspumontrealca          signal processing,
ste### [at] researchattcom           non stationnary signals
                                  and wavelets.
----------------------------------------------------------
         http://www.iro.umontreal.ca/~pigeon

Post a reply to this message

Cool!

A friend of mine compiled POV for Pentium Pro using MSVC5.0 max speed
optimisations. On a K6 ('cause I use one) it yields ~14% speed-up under NT
and ~5-6% under 95. What about your version? Anyway, I am downloading it
even now (5:50 a.m. local time!), but I was just curious if you can tell me
something beyond skyvase.

Regards,

---Peter

Post a reply to this message

Peter Popov wrote:

> Cool!
>
> A friend of mine compiled POV for Pentium Pro using MSVC5.0 max speed
> optimisations. On a K6 ('cause I use one) it yields ~14% speed-up under NT
> and ~5-6% under 95. What about your version? Anyway, I am downloading it
> even now (5:50 a.m. local time!), but I was just curious if you can tell me
> something beyond skyvase.
>
> Regards,
>
> ---Peter

What I found:

Parser is twice as fast.

Meshes are about 30% faster.

Average speed up is about 15% as reported by many
users... However, they all had Pentium 2's, and not K6's.

and:

-- Some minor bugs were corrected.
-- Some (minor) features were added, namely:
(in the currently available version on my web page)
   ETA counter, PPS with 2 digit precision (ex: 0.15 pps),
   pixel-in-line progress gadget,
(in the soon-to-be-on-my-web-page version)
   Wheel mouse support about everywhere, except in
   the editor window (because I can't compile it, my version
   of Delphi is too old), Render window refresh bug repaired
   (this bug was that sometimes the render window did not
   refresh correctly when another application changed the
   palette), ETA counter bug fix (eh... these things happens!),
   and... and... what else have I added or changed....?

Eventually, I'm going to profile the whole thing to see where CPU
time goes. Maybe I'll rewrite some small parts.

Best,

    S.

--
----------------------------------------------------------
Steven Pigeon                     Ph. D. Student.
University of Montreal.
pig### [at] iroumontrealca           Topics: data compression,
pig### [at] jspumontrealca          signal processing,
ste### [at] researchattcom           non stationnary signals
                                  and wavelets.
----------------------------------------------------------
         http://www.iro.umontreal.ca/~pigeon

Post a reply to this message

I rendered using the PII compile on a K6 200 it was more that 100%
slower on a 640x480 image of a bridge.
(Times were 3:23 for the optimized, 1:38 for the regular)

Steve


Peter Popov wrote:
> 
> Cool!
> 
> A friend of mine compiled POV for Pentium Pro using MSVC5.0 max speed
> optimisations. On a K6 ('cause I use one) it yields ~14% speed-up under NT
> and ~5-6% under 95. What about your version? Anyway, I am downloading it
> even now (5:50 a.m. local time!), but I was just curious if you can tell me
> something beyond skyvase.
> 
> Regards,
> 
> ---Peter

Post a reply to this message

Stephen Lavedas wrote:

> I rendered using the PII compile on a K6 200 it was more that 100%
> slower on a 640x480 image of a bridge.
> (Times were 3:23 for the optimized, 1:38 for the regular)

ohhh... that's bad. Are you sure that nothing else was running at
the same time? That's pretty weird... anyway, you're the first to
tell me that it slowed down, because everybody that tried it reported
speedups, not slowdowns.

What Peter Popov wrote:

> >> Cool!
> >
> > A friend of mine compiled POV for Pentium Pro using MSVC5.0 max speed
> > optimisations. On a K6 ('cause I use one) it yields ~14% speed-up under NT
> > and ~5-6% under 95. What about your version? Anyway, I am downloading it
> > even now (5:50 a.m. local time!), but I was just curious if you can tell me
> > something beyond skyvase.
> >
> > Regards,
> >
> > ---Peter

...is basically what I get from people who tries my version. Would you
do the experiments in the different order? Also, what is the scene? How
much memory do you have? Maybe one paid the penalty of getting stuff
into virtual memory...?

Or the K6 200 behaves really differently than the real Pentium 2 ?

Tell me more, I think this might be an important problem.


----------------------------------------------------------
Steven Pigeon                     Ph. D. Student.
University of Montreal.
pig### [at] iroumontrealca           Topics: data compression,
pig### [at] jspumontrealca          signal processing,
ste### [at] researchattcom           non stationnary signals
                                  and wavelets.
----------------------------------------------------------
         http://www.iro.umontreal.ca/~pigeon

Post a reply to this message

Steven Pigeon wrote:
> 
> Stephen Lavedas wrote:
> 
> > I rendered using the PII compile on a K6 200 it was more that 100%
> > slower on a 640x480 image of a bridge.
> > (Times were 3:23 for the optimized, 1:38 for the regular)
> 
> ohhh... that's bad. Are you sure that nothing else was running at
> the same time? That's pretty weird... anyway, you're the first to
> tell me that it slowed down, because everybody that tried it reported
> speedups, not slowdowns.
> 
I'm not sure.... I honestly don't remember, but the conditions were the
same for each render... I know that there were no ACTIVE windows as the
time... However, I must make on addition, the processor I'm using was
supposed to be a 233, but wouldn't run stably at 233 and I had to clock
it down.  I'll try it on my K6-300 at home this evening...

Steve

Post a reply to this message

>
>ohhh... that's bad. Are you sure that nothing else was running at
>the same time? That's pretty weird... anyway, you're the first to
>tell me that it slowed down, because everybody that tried it reported
>speedups, not slowdowns.
>
And I will be the second one. A julia fractal rendered 103% slower with the
*Pentium II* optimised version on my K6/200. I guess the conslusions are:

1. In the optimised version, all math is float (faster than int on P2,
slower on K6)
2. FPU operations are optimised for 2 floating point conveyors (K6 only has
one)
3. I'll stick to my K6 until I can afford a P2 (in a year maybe) and will
silently weep in a corner until then

BTW, in defense of AMD, a mandelbrot program a friend wrote in assembler and
manually optimised for the K6 was ~17% faster on AMD than Intel... Guess it
says it all about K6's RISC core and integer operations speed.

Post a reply to this message

Peter Popov <pet### [at] usanet> wrote in article
<35db94ca.0@news.povray.org>...
> >
> >ohhh... that's bad. Are you sure that nothing else was running at
> >the same time? That's pretty weird... anyway, you're the first to
> >tell me that it slowed down, because everybody that tried it reported
> >speedups, not slowdowns.
> >
> And I will be the second one. A julia fractal rendered 103% slower with
the
> *Pentium II* optimised version on my K6/200. I guess the conslusions are:
> 
<sniped rather obvious conclusions>
> 
> BTW, in defense of AMD, a mandelbrot program a friend wrote in assembler
and
> manually optimised for the K6 was ~17% faster on AMD than Intel... Guess
it
> says it all about K6's RISC core and integer operations speed.
> 

	Bollox does it ! Compare like against like (ie. K6 optimized on a K6 again
PII optimized on a PII) and then come back saying which is better. A K6
optimized app is always going to be better on a K6 then on an Intel chip,
just as a PII optimized app is always going to perform better on a PII than
any other chip. That's the point of optimization, it optimizes the code to
run the best on a given platform, you can't and shouldn't expect software
optimized for one specific platform to run equally efficiently on another,
different, platform.

-- 
Scott Hill
Sco### [at] DDLinkscouk
Software Engineer (and all round nice guy)
Company homepage : http://www.ddlinks.demon.co.uk

"The best trick the devil ever pulled was convincing people he didn't
exist..."
								- Verbal Kint.

"the Internet is here so we can waste time talking about nothing in 
 particular when we should be working" - Marcus Hill.

Post a reply to this message