OctaneRender® 1.024 beta 2.47 (lin/mac/win) [OBSOLETE]

Fri Jun 10, 2011 8:10 am

Hi, i have problems get 2.47 to work.
Cuda toolkit 4.0.17
Driver 270.41.06
Error message:

No cuda capable devices found, exiting

2.46b starts fine with cuda 3.2.16

Cheers, mib.

Fri Jun 10, 2011 8:21 am

Nicely done kubo! I'm wondering how your keeping the noise in the scene down as you have as many or more reflective surfaces in your render as in mine. I'm wondering, do you have any suggestions on how to deal with the hardwood floor issue I'm having trouble with?

kubo · Fri Jun 10, 2011 8:58 am

well, usually in my experience most crazy fireflies come from bump rather than specular materials, so getting them right (low enough so there is less fireflies but high enough that it's noticeable) usually does the trick, also normal maps work much better (there is a free nvidia pluging for photoshop that'll convert a b&w bump map in a normal) 'cause they give more detail with lower values. give it a try.

Fri Jun 10, 2011 9:03 am

Ah sorry kubo, I actually should have said normals. I'm currently using crazybump for them. Sory for the mixup. And really its not the fireflies but the overall noise of the reflections that giving me a hard time. Any other suggestions? Would there be much else adding noise to the floor?

kubo · Fri Jun 10, 2011 9:33 am

well, your noise just appears in the right side, where the floor is more "sided" could be a either "bump" related or also in some occasions with small and large sized light sources mixed in the image the small ones shed small noise, but with large samples (like I think you had) it should look smoother.

Fri Jun 10, 2011 9:43 am

And hence my conundrum. Makes me glad I decide to do this as a test before going to much further. I'm currently using a normal map that has small distortions in the wood other than the cracks. Could this be causing it?

kubo · Fri Jun 10, 2011 9:57 am

well, yes, that could be it, I thought it could be something like that, but in the right side under the window I couldn't see any relief, so I thought It was flat besides the table separations, but if it has any that's what's causing the "noise" which is not so, if you render that portion at higher resolution I'm sure you would see just the distortion product of the board's bump.

Fri Jun 10, 2011 10:06 am

I'll give it a shot and also try with/without the lights in the entry. I'll post comparisions as soon as I have them.

radiance · Fri Jun 10, 2011 11:13 am

Hi,

Thanks for reporting all the issues, we're looking into them (quite a lot of stuff to check)

Please continue testing and report your issues here.
I also think it might be good if some users can do some more tests with disperson+PMC, as not a lot of users have provided test renderings with it yet...

We have also nearly finished adding support for portals, some of you might already know what portals are as they are used in most unbiased engines,
you basically add a plane in an opening (such as a window) to guide the engine in sampling the environment light coming from an area outside of the area the portal is for,
which increases the convergence and efficiency a lot for interiors.

We might do another test build like this to test some fixes and the new portal feature next, it will depend on when the new multi-gpu framework is ready for testing,
which is also nearing completion. I can't make any promises but it won't be long, hopefully less than 2-3 weeks, as we've still some stuff to fix and some optimization work to do.

Regarding the sunlight through glass, I've decided not to fix this in 2.5, I know a lot of users are eager for a fix, but if we provide a fix for it now, with the current kernels,
we will waste quite a bit of development time, and the kernels will become slower as a result, while the addition of bidirectional path tracing will provide a solution to this problem as a welcome and effortless side-effect, amongst others.
Therefore I think it's better to bite the bullet, let users use the transparency workaround a bit longer, and work on bidirectional path tracing instead, which will take a bit longer,
but will be the best solution to the problem, and also will be the best and most adaptive/efficient kernel for most scenes.

Radiance

Fri Jun 10, 2011 11:32 am

radiance wrote:Hi,

Thanks for reporting all the issues, we're looking into them (quite a lot of stuff to check)
Please continue testing and report your issues here.
I also think it might be good if some users can do some more tests with disperson+PMC, as not a lot of users have provided test renderings with it yet...

We have also nearly finished adding support for portals, some of you might already know what portals are as they are used in most biased engines,
you basically add a plane in an opening (such as a window) to guide the engine in sampling the environment light coming from an area outside of the area the portal is for,
which increases the convergence and efficiency a lot for interiors.

We might do another test build like this to test some fixes and the new portal feature next, it will depend on when the new multi-gpu framework is ready for testing,
which is also nearing completion. I can't make any promises but it won't be long, hopefully less than 2-3 weeks, as we've still some stuff to fix and some optimization work to do.

Regarding the sunlight through glass, I've decided not to fix this in 2.5, I know a lot of users are eager for a fix, but if we provide a fix for it now, with the current kernels,
we will waste quite a bit of development time, and the kernels will become slower as a result, while the addition of bidirectional path tracing will provide a solution to this problem as a welcome and effortless side-effect, amongst others.
Therefore I think it's better to bite the bullet, let users use the transparency workaround a bit longer, and work on bidirectional path tracing instead, which will take a bit longer,
but will be the best solution to the problem, and also will be the best and most adaptive/efficient kernel for most scenes.

Radiance