Tweet
Hello,
We have uploaded a preview version of the next service pack, V-Ray 3.30.01, on our website here:
http://www.chaosgroup.com/en/2/downloads.html?s=v-ray
THESE ARE NOT THE FINAL BUILDS. We still have some work left on polishing this release, particularly for V-Ray RT GPU. We will probably post one more build before the release, probably in a couple of weeks.
However, for the moment I would like to get some feedback on a few of the new features, just to figure out if we are on the right track.
1) New adaptive sampling algorithm, both for the "Adaptive" bucket mode and the "Progressive" samplers. The goals of the new algorithm are:
a. To produce more even noise levels across the image. Our original algorithm tended to undersample darker parts of the image, while oversampling brighter parts.
b. To introduce a linear relationship between the noise threshold value and the render time. With the new sampler, decreasing the noise threshold twice always produces twice as many image samples and typically twice longer render times, while improving the quality respectively.
c. To better distinguish between areas with different variance. The old method could produce wildly varying noise levels under one and the same noise threshold value.
d. To always clean up a given area to the same noise level regardless of the actual underlying shading samples.
The new algorithm is not faster, it just produces images with more consistent quality, thereby reducing the need to fiddle with shading settings in materials and lights.
When loading an old scene, V-Ray will ask whether to switch to the new method or not; you can also switch manually with the following MaxScript command:
where N is 0 for the old method and 1 for the new method. New scenes use the new method by default. The switch option itself is not available in the UI.
2) Option to disregard shading subdivs in materials, lights and GI, with the minimum shading rate remaining the only shading quality control. This option can be toggled from the "DMC sampler" rollout. For new scenes, V-Ray by default disregards shading subdivs. For old scenes, V-Ray will ask whether to switch to the new default.
This means that by default, you have only one shading quality control - the minimum shading rate. We have done a lot of experiments with different scenes and various conditions to determine a good default value, which turned out to be 6. There is very little reason to change this value either up or down. In addition, V-Ray recognizes that some objects generally need higher AA (fur, hair, ForestPro) and automatically lowers the shading rate for these objects. The new adaptive sampling algorithm ensures that the final noise levels are still consistent.
My goal with this is to basically remove the need for the user to adjust any quality settings except for the image sampler. You can still do that if you like, of course, but from what I have seen, very few people manage to get better render times out of playing with the individual settings, and most people just make things worse. Unfortunately many optimziation tutorials on the internet just contribute to more confusion. V-Ray knows quite a bit about the rendered scene and it can actually allocate samples quite adequately on its own. Some more improvements are coming in this area too for the next builds.
3) Dynamic noise threshold for the progressive sampler. This is always enabled by default and there is very little reason to change it; still, it can be turned off from MaxScript with the command:
(the default value is 30).
This works by starting the progressive render with a high noise threshold and gradually reducing it until it gets to the specified value. The result is that the intermediate images have more consistent noise levels by causing the renderer to spend a bit more effort on the noisier parts first. In this way, the user has a better idea of the final image and if the rendering needs to be terminated after a certain time, makes sure that the best possible image for that time frame is calculated. If the render is not limited in time, it eventually converges to the same final image as when the option is disabled - just the way to get there is different.
The difference for many scenes is not actually that drastic, but it does produce images of more uniform quality while the progressive sampler is working. Also, V-Ray will print out the currently reached noise threshold in the V-Ray messages window so that you can reproduce the same quality later on if needed.
4) Updated sun and sky models (Hosek et al model+ground albedo parameters) plus aerial perspective environment effect. This build also changes the internal spectral RGB color space of V-Ray to sRGB from CIE RGB which we used in previous builds. This affects calculations of the sun and sky, as well as any conversions of Kelvin temperature to RGB colors. For new scenes, the sRGB color space is used. This can be toggled with the MaxScript command:
where N is 1 for sRGB, and 0 for the old CIE RGB.
The updated color space has two effects:
a. The V-Ray sun and sky have slightly warmer colors rather than the greenish tint they had before;
b. The white point of this color space is 6500K, meaning that lights at 6500K and camera white balance at 6500K produce the neutral white result.
5) Raytraced rounded corners. This is enabled by default both for old and new scenes; it can be disabled by setting the "roundedCorners_raytraced" parameter of VRayEdgesTex to "false". This mode enables the rounded corners algorithm to:
a. Work better in general, by producing smoother results with less artifacts;
b. Be able to blend normals between different objects;
c. To work with displaced/subdivided objects, as well as VRayProxy objects;
d. To work with disconnected meshes.
Currently this works on the CPU only; we do have an implementation for V-Ray RT GPU, but it wasn't quite ready for these builds. I'm sure the question of include/exclude objects will come up, as well as an additional bump slot inside the VRayEdgesTex texture itself, and these are planned. In case you are wondering, we licensed the raytraced rounded corners method from nVidia, although we had to improve the original algorithm a bit in order to make it useful. I think that it works better than the mental ray/iray implementations.
There are also many other improvements that you can see listed in the changelog below.
Any feedback is welcome!
Best regards,
Vlado
We have uploaded a preview version of the next service pack, V-Ray 3.30.01, on our website here:
http://www.chaosgroup.com/en/2/downloads.html?s=v-ray
THESE ARE NOT THE FINAL BUILDS. We still have some work left on polishing this release, particularly for V-Ray RT GPU. We will probably post one more build before the release, probably in a couple of weeks.
However, for the moment I would like to get some feedback on a few of the new features, just to figure out if we are on the right track.
1) New adaptive sampling algorithm, both for the "Adaptive" bucket mode and the "Progressive" samplers. The goals of the new algorithm are:
a. To produce more even noise levels across the image. Our original algorithm tended to undersample darker parts of the image, while oversampling brighter parts.
b. To introduce a linear relationship between the noise threshold value and the render time. With the new sampler, decreasing the noise threshold twice always produces twice as many image samples and typically twice longer render times, while improving the quality respectively.
c. To better distinguish between areas with different variance. The old method could produce wildly varying noise levels under one and the same noise threshold value.
d. To always clean up a given area to the same noise level regardless of the actual underlying shading samples.
The new algorithm is not faster, it just produces images with more consistent quality, thereby reducing the need to fiddle with shading settings in materials and lights.
When loading an old scene, V-Ray will ask whether to switch to the new method or not; you can also switch manually with the following MaxScript command:
Code:
renderers.current.twoLevel_adaptiveMethod=N
2) Option to disregard shading subdivs in materials, lights and GI, with the minimum shading rate remaining the only shading quality control. This option can be toggled from the "DMC sampler" rollout. For new scenes, V-Ray by default disregards shading subdivs. For old scenes, V-Ray will ask whether to switch to the new default.
This means that by default, you have only one shading quality control - the minimum shading rate. We have done a lot of experiments with different scenes and various conditions to determine a good default value, which turned out to be 6. There is very little reason to change this value either up or down. In addition, V-Ray recognizes that some objects generally need higher AA (fur, hair, ForestPro) and automatically lowers the shading rate for these objects. The new adaptive sampling algorithm ensures that the final noise levels are still consistent.
My goal with this is to basically remove the need for the user to adjust any quality settings except for the image sampler. You can still do that if you like, of course, but from what I have seen, very few people manage to get better render times out of playing with the individual settings, and most people just make things worse. Unfortunately many optimziation tutorials on the internet just contribute to more confusion. V-Ray knows quite a bit about the rendered scene and it can actually allocate samples quite adequately on its own. Some more improvements are coming in this area too for the next builds.
3) Dynamic noise threshold for the progressive sampler. This is always enabled by default and there is very little reason to change it; still, it can be turned off from MaxScript with the command:
Code:
renderers.current.progressive_dynNoiseThreshold=0
This works by starting the progressive render with a high noise threshold and gradually reducing it until it gets to the specified value. The result is that the intermediate images have more consistent noise levels by causing the renderer to spend a bit more effort on the noisier parts first. In this way, the user has a better idea of the final image and if the rendering needs to be terminated after a certain time, makes sure that the best possible image for that time frame is calculated. If the render is not limited in time, it eventually converges to the same final image as when the option is disabled - just the way to get there is different.
The difference for many scenes is not actually that drastic, but it does produce images of more uniform quality while the progressive sampler is working. Also, V-Ray will print out the currently reached noise threshold in the V-Ray messages window so that you can reproduce the same quality later on if needed.
4) Updated sun and sky models (Hosek et al model+ground albedo parameters) plus aerial perspective environment effect. This build also changes the internal spectral RGB color space of V-Ray to sRGB from CIE RGB which we used in previous builds. This affects calculations of the sun and sky, as well as any conversions of Kelvin temperature to RGB colors. For new scenes, the sRGB color space is used. This can be toggled with the MaxScript command:
Code:
renderers.current.options_rgbColorSpace=N
The updated color space has two effects:
a. The V-Ray sun and sky have slightly warmer colors rather than the greenish tint they had before;
b. The white point of this color space is 6500K, meaning that lights at 6500K and camera white balance at 6500K produce the neutral white result.
5) Raytraced rounded corners. This is enabled by default both for old and new scenes; it can be disabled by setting the "roundedCorners_raytraced" parameter of VRayEdgesTex to "false". This mode enables the rounded corners algorithm to:
a. Work better in general, by producing smoother results with less artifacts;
b. Be able to blend normals between different objects;
c. To work with displaced/subdivided objects, as well as VRayProxy objects;
d. To work with disconnected meshes.
Currently this works on the CPU only; we do have an implementation for V-Ray RT GPU, but it wasn't quite ready for these builds. I'm sure the question of include/exclude objects will come up, as well as an additional bump slot inside the VRayEdgesTex texture itself, and these are planned. In case you are wondering, we licensed the raytraced rounded corners method from nVidia, although we had to improve the original algorithm a bit in order to make it useful. I think that it works better than the mental ray/iray implementations.
There are also many other improvements that you can see listed in the changelog below.
Any feedback is welcome!
Best regards,
Vlado
We put the builds mostly for those that have some time to experiment and provide feedback. While the builds have been used in production, if you have current projects it may be better to wait for the official release in a month.
Comment