[Sidefx-houdini-list] stability of mantra on multi core machines?

Nicholas Pliatsikas nick_p102 at hotmail.com
Tue Aug 5 06:31:46 EDT 2008


We have started to see similar stable results from mantra as whole also. Early on in the 9 and 9.1 dev there was alot of crashing when using mantra with mutli cores. That seems to be sorted now thankfully. We where using 9.1.239 mostly. From the latest project I worked on it looks to be that our main issues were first and foremost human error :¬D after that mantra related issues where memory usage with instances Delayed read archives and motionblur created memory explosions, again this had been solved in the new builds.
 
In general memory usage is pretty good, with standard scenes not hitting more then a gig of ram, on certain scenes we did have ram usage up to 8 gigs which would require all 4 cores. We tended to use multple jobs per machine when ram wasn't hitting limits i.e 2 gig limits per core. One mantra problem that did affect my work and only happened the once, my renders would stop at specific point every time. We managed to figure out that mantra seemed to hit an infinite loop and tried to access the same textures over and over. The strange thing was that the two elements causing this problem would render by them selves, but not together no matter what we tried. Turn one of the offending objects to matte and did a separate pass to solve the issue. 
 
We did how ever find generating IFDs consistently was more of a problem on our farm. If mantra didn't render its usually due to a busted IFD. The IFD gen sometimes easily hit over 8 gigs on complex scenes Especially in regards to heavy use of grass guides and embedded geometry i.e not DRAs.
 
Mantras been pretty great with what we have chucked at it. Most other rendering solutions would have fallen over, probably with the exception of renderman since the workings are quite similar.
 
Nicholas Pliatsikas
 



> Date: Mon, 4 Aug 2008 16:15:14 -0700> From: jiversen at rhythm.com> To: sidefx-houdini-list at sidefx.com> Subject: Re: [Sidefx-houdini-list] stability of mantra on multi core machines?> > We've just come off The Mummy3 where all the FX elements are rendered > solely in Mantra.> > We settled on 9.1.179 during the production and I was quite amazed that > we didn't experience a single unexplained crash or deadlock from Mantra > for the entire production (that I know of)... something I've never had > the pleasure of experiencing with any other renderer (yes, even with > PRMan) or production.> > We did blow our RAM a lot as we were rendering huge volumetric scenes > and our procedural was holding onto those voxel caches which took a > chunk of memory. I, too, would love it if Mantra could use a memory > manager to try to handle RAM limits more gracefully - but in a couple of > years I suppose that everything will be 64bit and this shouldn't matter > much anymore... and I'd rather push other features than this.> > Michael, I think you must either be on a bad build or perhaps have a bad > DSO or something because I feel very warm about Mantra's stability now > more than ever. Which version are you running on? Moreover, any issues > I've reported the past (with an IFD for reproducible results) have been > fixed with a very short turnaround.> > > Michael K O'Brien wrote:> > Hola~> >> > Yes. If the render cores out, it's pretty easy to detect that and retry the> > frame. However, it constantly just deadlocks. I can't imagine trying to use> > mantra on a production.> >> > MO> >> > --> > Michael K O'Brien> > mobrien at pixar.com> > Pixar> >> > JEN: Just because it's Taco Bell doesn't mean it's not dinner!> >> > > >> -----Original Message-----> >> From: sidefx-houdini-list-bounces at sidefx.com [mailto:sidefx-houdini-list-> >> bounces at sidefx.com] On Behalf Of Simon Kapeniak> >> Sent: Tuesday, July 29, 2008 12:49 AM> >> To: sidefx-houdini-list at sidefx.com> >> Subject: Re: [Sidefx-houdini-list] stability of mantra on multi core> >> machines?> >>> >> The question is why 8 cores rendering eats much more memory then 1 core.> >> Don't cores share memory? I understand 20% differences, but Mantra likes> >> to> >> take as big as twice ram in mulithreding mode. It also doesn't explain> >> the> >> situation in which mantra doesn't fail but freezes on a few buckets in> >> multi-core rendering.> >>> >> 2008/7/29 Pablo Giménez <pablogipi at gmail.com>> >>> >> > >>> Yes unfortunately, when you reach the memory limit of your machines you> >>> > >> get> >> > >>> random fails, and mantr exists only with a segfault, very bad.> >>> I has a similar problems some months ago and I rememberd struggling> >>> > >> myself> >> > >>> trying to find the problem and it was the memory, nothing else, but> >>> > >> because> >> > >>> Mantra exists without any info you can waste a lot of time trying to> >>> > >> figure> >> > >>> out what is happening.> >>> And also, rendering with 4 cores seems pretty solid, with eith cores I> >>> > >> have> >> > >>> experimented problems in early versions if mantra 9 and 9.1.> >>> Cheers> >>>> >>> 2008/7/28 paul simpson <paul at realisestudio.com>> >>>> >>> > >>>> for linux: man ulimit> >>>>> >>>> 2008/7/28 Guillaume Laforge <guillaume at alamaison.fr>> >>>>> >>>> > >>>>> We've got the memory limit problem here when rendering fur> >>>>> > >> procedurals> >> > >>> !> >>> > >>>>> If the memory is closed to the limit, frames randomly crash.> >>>>> As a workaround we set a time bomb on our rendermanager to re-render> >>>>> those bad frames. The same frames can crash twice or more and then> >>>>> render fine.> >>>>> I don't know if there is a way to set a memory limit in Mantra as in> >>>>> Mental Ray ?> >>>>>> >>>>> Cheers> >>>>>> >>>>> Guillaume Laforge | la maison <http://www.alamaison.fr>> >>>>> my new little blog : http://frenchdog.wordpress.com/> >>>>> <http://frenchdog.wordpress.com/>> >>>>>> >>>>>> >>>>> Guillaume Laforge a écrit :> >>>>> > >>>>>> _______________________________________________> >>>>>> 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> >>> _______________________________________________> >>> 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> > > > -- > jason iversen> fx supervisor> r+h> > is this awesome? (y/n)> > > _______________________________________________> Sidefx-houdini-list mailing list> Sidefx-houdini-list at sidefx.com> https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
_________________________________________________________________
Win New York holidays with Kellogg’s & Live Search
http://clk.atdmt.com/UKM/go/107571440/direct/01/


More information about the Sidefx-houdini-list mailing list