I'm trying to construct some space scenery containing a few very large
objects (Sun, Earth, Moon), and a small number of space vessels (~200 m in
length).  I started out using a scale of 1 unit = 1 m, but found that large
objects in the scene disappeared - they simply weren't being rendered.  I
switched to 1 unit = 1e6 m, which I thought might be a good compromise,
which is better, but the image is very grainy when both large and small
objects are rendered in the same scene.  I think these effects must be due
to limits of floating-point precision (this is POV-Ray 3.6 on 32-bit
Windows XP).  Is there any kind of workaround for this?  I'd really like to
work to scale if possible.

Is the 64-bit version of POV-Ray likely to work better for this kind of
scenery? I am using an AMD64 system but currently have no 64-bit Windows
installation; IIRC, SSE/SSE2 has 128-bit FP registers.

Thanks in advance!

Post a reply to this message

"screwtop" <cme### [at] ihugconz> schreef in bericht
news:web.440cd3ff38a1658e9f9718b50@news.povray.org...
> I'm trying to construct some space scenery containing a few very large
> objects (Sun, Earth, Moon), and a small number of space vessels (~200 m in
> length).  I started out using a scale of 1 unit = 1 m, but found that
large
> objects in the scene disappeared - they simply weren't being rendered.  I
> switched to 1 unit = 1e6 m, which I thought might be a good compromise,
> which is better, but the image is very grainy when both large and small
> objects are rendered in the same scene.  I think these effects must be due
> to limits of floating-point precision (this is POV-Ray 3.6 on 32-bit
> Windows XP).  Is there any kind of workaround for this?  I'd really like
to
> work to scale if possible.
>
> Is the 64-bit version of POV-Ray likely to work better for this kind of
> scenery? I am using an AMD64 system but currently have no 64-bit Windows
> installation; IIRC, SSE/SSE2 has 128-bit FP registers.
>

I am not sure if this will help you, but have a look at this message:
http://news.povray.org/436572fd%40news.povray.org
and the few following that (or even the initial discussion). You would need
version 3.6.0. Can be found on the ftp site of POV-Ray I think, or one of
the mirror sites.

Thomas

Post a reply to this message

"Thomas de Groot" <t.d### [at] internlnet> wrote:
> "screwtop" <cme### [at] ihugconz> schreef in bericht
> news:web.440cd3ff38a1658e9f9718b50@news.povray.org...
> > I'm trying to construct some space scenery containing a few very large
> > objects (Sun, Earth, Moon), and a small number of space vessels (~200 m in
> > length).  I started out using a scale of 1 unit = 1 m, but found that
> large
> > objects in the scene disappeared - they simply weren't being rendered.  I
> > switched to 1 unit = 1e6 m, which I thought might be a good compromise,
> > which is better, but the image is very grainy when both large and small
> > objects are rendered in the same scene.  I think these effects must be due
> > to limits of floating-point precision (this is POV-Ray 3.6 on 32-bit
> > Windows XP).  Is there any kind of workaround for this?  I'd really like
> to
> > work to scale if possible.
> >
> > Is the 64-bit version of POV-Ray likely to work better for this kind of
> > scenery? I am using an AMD64 system but currently have no 64-bit Windows
> > installation; IIRC, SSE/SSE2 has 128-bit FP registers.
> >
>
> I am not sure if this will help you, but have a look at this message:
> http://news.povray.org/436572fd%40news.povray.org
> and the few following that (or even the initial discussion). You would need
> version 3.6.0. Can be found on the ftp site of POV-Ray I think, or one of
> the mirror sites.
>
> Thomas

Hello!

I already encountered this kind of problem. I surely comes from POV's
internal epsilons (as shown in the above-mentioned thread). So one must
think of designing scenes with reasonable range for the distances (eg
billion miles away planet/star vs milimetric objects in the foreground)

Perhaps some epsilons could re-adjusted in 3.7 (I don't know much about
POV's internals). In addition, I don't think processor-precision has much
to do with this behaviour compared to these epsilons. Maybe someone more
informed about POV's intimacy could enlighten us. Thanks to him in advance
:)

   Bruno

Post a reply to this message

"Bruno Cabasson" <bru### [at] alcatelaleniaspacefr> wrote:
> Perhaps some epsilons could re-adjusted in 3.7 (I don't know much about
> POV's internals). In addition, I don't think processor-precision has much
> to do with this behaviour compared to these epsilons. Maybe someone more
> informed about POV's intimacy could enlighten us. Thanks to him in advance
As a result of the above-mentioned thread (and another that followed on), it
was my understanding that the enhanced precision will indeed be included in
3.7. The powers that be can confirm / deny this, I'm sure.

Bill

Post a reply to this message

screwtop nous apporta ses lumieres en ce 06/03/2006 19:31:
> I'm trying to construct some space scenery containing a few very large
> objects (Sun, Earth, Moon), and a small number of space vessels (~200 m in
> length).  I started out using a scale of 1 unit = 1 m, but found that large
> objects in the scene disappeared - they simply weren't being rendered.  I
> switched to 1 unit = 1e6 m, which I thought might be a good compromise,
> which is better, but the image is very grainy when both large and small
> objects are rendered in the same scene.  I think these effects must be due
> to limits of floating-point precision (this is POV-Ray 3.6 on 32-bit
> Windows XP).  Is there any kind of workaround for this?  I'd really like to
> work to scale if possible.
> 
> Is the 64-bit version of POV-Ray likely to work better for this kind of
> scenery? I am using an AMD64 system but currently have no 64-bit Windows
> installation; IIRC, SSE/SSE2 has 128-bit FP registers.
> 
> Thanks in advance!
> 
> 
> 
> 
Look at the other replys first.
As for your CPU, 32 bits and 64 bits both have 64 bits FPU. In fact, even the latest
16 bits CPU, 
those with integrated FPU, where equiped with 64 bits FPU. 64 bits is the size
required for double 
presision floating point operations.
If you have the source code, you can change the epsilon value and compile it, but be
carefull not to 
overdo it. As the values get smaler, you can get very close to the resolution limit
imposed by the 
double presision operations. This, in turn, will lead to more rounding and
granulausity errors, 
causing "random" artefacts.

-- 
Alain
-------------------------------------------------
BREAKFST.COM halted . . . cereal port not responding!

Post a reply to this message