pouët.net

Go to bottom

rendering with cuda

category: general [glöplog]
 

been looking into cuda and tesla as gpu rendering solutions for max or maya rendering. seems that nvidia bought the company behind mentalray some 9 months ago or so, so rumours about cuda and tesla being adapted as plugin for traditional software rendering programs are aplenty, but gellato was dropped/discontinued/opensourced/whatever and amaretto feels abit too beta (and now officially unsupported) to consider for seirous use on max rendering projects.

it still feels abit retro to use cpu based cluster/renderfarms in 2009 though (despite multicore and all that jazz, so, you folks who might be more into this kind of stuff, are there any decent gpu based rendering alternatives out there that i might not be aware of? integrated into max preferably.
added on the 2009-03-12 13:30:12 by psenough psenough
no renderfarm folks reading pouet?
added on the 2009-03-14 00:54:27 by psenough psenough
there's RTSquare for 3ds max (www.rtsquare.eu IIRC). It has a trial version you might want to check)... about it's performance or even if it's any good, I don't really know :-)
added on the 2009-03-14 02:50:01 by Jcl Jcl
duh! I missed the Amaretto thing :D

BB Image

BB Image

added on the 2009-03-14 03:36:05 by bdk bdk
disaronno :Q___
added on the 2009-03-14 10:50:44 by rmeht rmeht
jcl: i had run into that info before, that page seems discontinued for a few years now though (which made me wonder about the product itself)
added on the 2009-03-14 11:42:05 by psenough psenough
ask here instead.
added on the 2009-03-14 12:05:49 by bonzaj bonzaj
are the GPU calculations IEEE accurate already?
added on the 2009-03-14 13:14:15 by the_Ye-Ti the_Ye-Ti
theyeti: http://en.wikipedia.org/wiki/CUDA

Quote:
For double precision there are no deviations from the IEEE 754 standard. In single precision, Denormals and signalling NaNs are not supported; only two IEEE rounding modes are supported (chop and round-to-nearest even), and those are specified on a per-instruction basis rather than in a control word (whether this is a limitation is arguable); and the precision of division/square root is slightly lower than single precision.


i guess so. atleast there are shitloads of serious simulations and visualization systems using tesla / cuda already, doubt they'd use it for medical visualization systems if there were float precision problems but i might be wrong.
added on the 2009-03-14 15:04:29 by psenough psenough
Quote:
ask here instead.


didnt know about that forum, thanks for the tip.
added on the 2009-03-14 15:12:01 by psenough psenough
Just how much noise are you planning on rendering so that it requires a renderfarm? </LOL>
ompf forum thread trackback, in case anyone else is interested on following it up.
added on the 2009-03-14 15:27:05 by psenough psenough

login

Go to top