On 10/01/2011 03:22, stevenvh wrote:
> Hi,
> I'm trying to render an image at 19050x7620 resolution in v3.7 RC2. The image
> has radiosity, so the rendering takes some iterations. After some 40% of the
> last iteration (pixel level) it stops with an "out of memory" error.
> Is there a way to increase memory, like by some parameter in an ini-file, so I
> can render the image completely?
> TIA
> Steven
> 
> I'm running v3.7 RC2 under Windows XP on a 4GB RAM machine.

RC2 has a memory indicator in the status bar just to the right of the
messages. Did you happen to notice what that was showing during the render
(especially towards the end?).

I probably ought to add code to print the memory stats even if a render
fails; currently, they're only printed if it succeeds. In cases like this
it would definitely have been worth having this info as I can't say without
knowing how much of your machine's physical resources it was using.

-- Chris

Post a reply to this message

Chris Cason <del### [at] deletethistoopovrayorg> wrote:
>
> RC2 has a memory indicator in the status bar just to the right of the
> messages. Did you happen to notice what that was showing during the render
> (especially towards the end?).
>
> I probably ought to add code to print the memory stats even if a render
> fails; currently, they're only printed if it succeeds. In cases like this
> it would definitely have been worth having this info as I can't say without
> knowing how much of your machine's physical resources it was using.
>


Thanks for the reply, Chris.

No, I never took note of the memory indicator before. Anyway, this render had
been running for 6+ hours, and I wasn't always near my PC.
I'll restart the render tomorrow and try to monitor it.

I agree about printing the memory stats. I would think it's of little use if
everything went well, but be especially useful in case of errors.

Post a reply to this message

The .pov-state file after the error is 1.2GB. Is that an indication?

Post a reply to this message

On 10/01/2011 05:25, stevenvh wrote:
> The .pov-state file after the error is 1.2GB. Is that an indication?

no, unfortunately. however, do keep in mind that you can resume a render
using that file (radiosity info that was not saved separately would need to
be re-calculated though).

-- Chris

Post a reply to this message

I captured the complete error message from yesterday's crash. I thought this
might be interesting as well:

**************************************************
Memory allocation failure exception thrown at base\povmscpp.cpp line 1908.
POVMSAttr_Set failed, out of memory
Memory allocation failure exception thrown at base\povmscpp.cpp line 1908.
Fatal error in renderer: Out of memory.
Render failed
**************************************************

I managed to trace today's rendering until the fatal error. I get the same
errors except that the POVMSAttr_Set line is missing.

Memory use:
In the last iteration (pixel level) stable at around 980MB until 25% is
rendered, then it starts decreasing until it is 560MB just before fatal error.
After it indicates 430MB (FWIW).

I'm running this on a quad-core Phenom 9500, which to my surprise was only ~18%
loaded, load spread evenly over the 4 cores. Rest of the time idle. Shouldn't
POV-Ray use close to 100% of cycles?

Post a reply to this message

The scene in my test had 71000 objects. To know if the number of objects is
relevant in this or not, I retried with just simple 3 objects. The result is the
same, at the exact same position: 43% in the final iteration.
Are there people who successfully managed to render high resolution images in
RC2, in the order of 100 million pixels?

Post a reply to this message

Le 10/01/2011 18:31, stevenvh nous fit lire :
> Are there people who successfully managed to render high resolution images in
> RC2, in the order of 100 million pixels?
> 
> 
> 

Yes.
ovus.png PNG 19050x7620 19050x7620+0+0 8-bit DirectClass 5.659MB

Finite Objects:            4
Infinite Objects:          0
Light Sources:             1
Total:                     5

Done with Povray 3.7RC2 on Ubuntu(10.10) 64 bits, with enough disk & memory.
 ( povray ovus.pov +W19050 +H7620 +RP0 +RS1 -A )

Note: Despite the number of threads (-WT) by default, after a quick
start, there is mainly one core at 100% and the others off, due probably
to I/O to disk.

Post a reply to this message

Le_Forgeron <jgr### [at] freefr> wrote:
> Le 10/01/2011 18:31, stevenvh nous fit lire :
> > Are there people who successfully managed to render high resolution images in
> > RC2, in the order of 100 million pixels?
> >
> >
> >
>
> Yes.
> ovus.png PNG 19050x7620 19050x7620+0+0 8-bit DirectClass 5.659MB
>
> Finite Objects:            4
> Infinite Objects:          0
> Light Sources:             1
> Total:                     5
>
> Done with Povray 3.7RC2 on Ubuntu(10.10) 64 bits, with enough disk & memory.
>  ( povray ovus.pov +W19050 +H7620 +RP0 +RS1 -A )
>

Merci, Le Forgeron.
Maybe the fact that you can complete the rendering has to do with yours being
the 64-bit version, while mine is 32-bits.

Post a reply to this message

Am 10.01.2011 21:48, schrieb Le_Forgeron:

> Finite Objects:            4
> Infinite Objects:          0
> Light Sources:             1
> Total:                     5
>
> Note: Despite the number of threads (-WT) by default, after a quick
> start, there is mainly one core at 100% and the others off, due probably
> to I/O to disk.

Yes, I guess that's to be expected with a pretty simple scene at high 
resolutions: Storing a pixel into the image buffer seems to be taking 
more time than computing it.

Post a reply to this message

Le 11/01/2011 18:45, clipka nous fit lire :
>> Note: Despite the number of threads (-WT) by default, after a quick
>> start, there is mainly one core at 100% and the others off, due probably
>> to I/O to disk.
> 
> Yes, I guess that's to be expected with a pretty simple scene at high
> resolutions: Storing a pixel into the image buffer seems to be taking
> more time than computing it.

I hope you won't mind that I change it a bit to get back some real
performance (at the cost of a bit of memory, but using only one core for
I/O when many were sitting doing nothing was just not my taste: now
everyone is working if you do not render in too fancy order...)
64 lines of cache instead of one seems a small price to my happiness.
(Ovus at 19000+ render now in 35 seconds, and about the same time to
write the PNG, instead of taking many many minutes of slow rendering)
 ( 373.849 CPU-seconds total, but the I/O were really killing the rendering)

I know it's RC2, but I think #5382 might get in, really (please review &
check)

Post a reply to this message