[Sidefx-houdini-list] 8 core render hickups

Pablo Giménez pablogipi at gmail.com
Tue Mar 25 07:43:32 EDT 2008


2008/3/24, Jed Schwartz <jed at jedschwartz.com>:
>
> Just a quick note to confirm setting thread limit to 4 and spreading
> the renders out over a couple of processes did the trick- We rendered
> all weekend and never missed a frame-


I remember having problems with multithreading and raytracing renders some a
couple of months ago.
Putting the render to 4 threads only dis the trick too, when I tried to put
it to 8 I get problems or even the render is slower compared with 4 threads.
I sent some emails to SESI and they said that there ar some problems with
multithreading in Mantra, so I suppose that some of these problems stills
there.

Thanks Guys!
>
> Jed-
>
>
> On Mar 22, 2008, at 10:46 AM, paul simpson wrote:
>
> > for what it's worth - we've been rendering 1000's of frames
> > recently with
> > 8cores per frame (using 5-6G RAM).  andrew has just fixed some
> > random thread
> > related crashes - so get a recent build - cos that could well help.
> >
> > we have certainly found that mantra exists disgracefully if it runs
> > out of
> > ram.  all our frames started going though very smoothly once we
> > went from
> > 4cores/4G to 8 cores/8G.
> >
> >
> > On 22/03/2008, Simon Kapeniak <szymon.kapeniak at gmail.com> wrote:
> >>
> >> How many threads you use per frame? There were quit a few reports
> >> (H9.0and
> >> H9.1) of random crashed in multithreading mode. Specially during
> >> rendering
> >> many frames (statistical relation probably). If RAM allows tuning
> >> mantra
> >> down to 2-4 threads per frame and splitting job in two processes
> >> could
> >> help
> >> a lot. Although 1 broken frame per a 100 is not so bad I suppose.
> >> Better
> >> than out-of-ram issue. I don't have many experiences with PRMan, but
> >> mental
> >> ray with its renowned multithreading makes the same problem.
> >> Buckets are
> >> loosing their data in random frames etc. Not sure if these sort of
> >> problem
> >> can be eradicated fully anyway.
> >>
> >> cheers,
> >> Simon.
> >>
> >>
> >>
> >> 2008/3/22, Jed Schwartz <jed at jedschwartz.com>:
> >>
> >>>
> >>> Hey Peter,
> >>>
> >>> We are using Ubuntu 64-- Memory does not seem to be the problem-
> >>> Monitoring processes, I can see that Ubuntu  allows way over 2.8
> >>> gigs
> >>> of the (8 gigs) available ram to be used during a "heavy" render
> >>> without any problem-  I'm still getting sporadic interruptions in
> >>> renders- Sometimes from within the package, and sometimes processing
> >>> ifd's from command line- Again, not a major inconvenience so far,
> >>> Been too busy with production to really focus on tracking down the
> >>> problem- Any suggestions where to look will be much appreciated!
> >>>
> >>> Thanks!
> >>>
> >>>
> >>> Jed-
> >>>
> >>>
> >>>
> >>> On Mar 21, 2008, at 11:34 AM, Peter Bowmar wrote:
> >>>
> >>>> Hey Jed,
> >>>>
> >>>> Are you using 64bit Ubuntu and Houdini builds? If not, doesn't
> >>>> matter
> >>>> how much RAM you have, 2.8gb is your limit... Also, remember
> >>>> that even
> >>>> though Mantra 9's threading uses shared memory, using more threads
> >>>> will still use some more memory because you'll have multiple
> >>>> buckets
> >>>> being processed simultaneously. Try a single thread with the
> >>>> smallest
> >>>> bucket size as the first step in testing if memory is an issue.
> >>>> Also
> >>>> watch "top" and see if Mantra is hitting 2.8gb or so.
> >>>>
> >>>> Cheers,
> >>>>
> >>>> Peter B
> >>>>
> >>>> On 10/03/2008, Jed Schwartz <jed at jedschwartz.com> wrote:
> >>>>> We set up a new 8 core MacPro running ubuntu 7.10 a couple of
> >>>>> weeks
> >>>>>  ago- The system is blazingly fast and quite wonderful to work on.
> >>>>>  The only problem we're having is scattered segmentation fault
> >>>>> errors
> >>>>>  when rendering-
> >>>>>
> >>>>>  When rendering from within the package, or rendering a batch of
> >>>>>  indexed .ifd files from the command line, once or twice in 100
> >>>>> frames
> >>>>>  I'm getting "segmentation fault (core dumped)"  - and the
> >>>>> frame does
> >>>>>  not render- - Occasionally the render stops, but usually  my
> >>>>> script
> >>>>>  continues the render without problems-
> >>>>>
> >>>>>   When I monitor processes I can see memory is holding up fine (8
> >>>>>  megs) and all 8 processors are working away on the frames- It's a
> >>>>>  minor irritation to pick up the few missing frames, bit I'm
> >>>>> thinking
> >>>>>  that I must be missing something in my ROP that's causing this
> >>>>>  behavior-?
> >>>>>
> >>>>>  Anyone have any idea of what's going on-?
> >>>>>
> >>>>>  Thanks!
> >>>>>
> >>>>>  Jed-
> >>>>>
> >>>>>  _______________________________________________
> >>>>>  Sidefx-houdini-list mailing list
> >>>>>  Sidefx-houdini-list at sidefx.com
> >>>>>  https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> >>>>>
> >>>> _______________________________________________
> >>>> Sidefx-houdini-list mailing list
> >>>> Sidefx-houdini-list at sidefx.com
> >>>> https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> >>>
> >>> _______________________________________________
> >>> Sidefx-houdini-list mailing list
> >>> Sidefx-houdini-list at sidefx.com
> >>> https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> >>>
> >> _______________________________________________
> >> Sidefx-houdini-list mailing list
> >> Sidefx-houdini-list at sidefx.com
> >> https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> >>
> > _______________________________________________
> > Sidefx-houdini-list mailing list
> > Sidefx-houdini-list at sidefx.com
> > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
>
> _______________________________________________
> Sidefx-houdini-list mailing list
> Sidefx-houdini-list at sidefx.com
> https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
>



-- 

Un saludo
Best Regards
Pablo Giménez



More information about the Sidefx-houdini-list mailing list