Dome Light resolution 8K?

[dan_plattner]

Hi all!
I wonder, if I use a Dome light to lit the scene and the Dome Light supports only up to 8K resolution, so all the HDR files that are larger than 8K are useless in term of resolution?
Or in other words, the Dome Light can't take advantage of HDR files that are larger than 8K because the dome light compress the files?

[joconnell]

There's one resolution for the light calculations and another for reflections and refractions. For lighting, lower resolution means cleaner results in terms of noise so you can often get a lot of speed benefits by using smaller numbers here. For a backplate or for the reflections in an image, vray will use full quality.

[^Lele^]

Hi all!
I wonder, if I use a Dome light to lit the scene and the Dome Light supports only up to 8K resolution, so all the HDR files that are larger than 8K are useless in term of resolution?
Or in other words, the Dome Light can't take advantage of HDR files that are larger than 8K because the dome light compress the files?

The resolution you see in the dome light controls (marked as "Res:") governs the bitmap importance sampling resolution.
That's the process by which V-Ray figures out which parts of the input image will contribute the most to lighting (f.e., in an outdoor daylight probe, it will find the Sun).
The Importance Sampled image will then be used for the GI calculations, while the untouched input image will be used for reflections and refractions.

It's important to note that this is *not* a simple downsampling, or resizing of the input image.
Rather, you can think of the resolution of it more as the number of partitions into which the original image will be divided into so to find the brightest parts.
The resulting Importance Sampled image will have all the relevant data it'll need from the input, but require (a lot!) less sampling overall to converge to a clean solution.

Normally you shouldn't need to change the resolution at all, no matter the input, from its default of 2048 (at rendertime any lower resolution will be bumped to that value.).
There may be some extreme cases (synthetic in nature, surely.) where raising the value may add some detail to the lighting, but i am still working at creating one, so far without much success.

I'll post some comparisons as soon as they are ready.

[kosso_olli]

Lele, does this value have any effect on render speed or noise? If so, what resolution would be best for an outdoor HDRI with the sun as main light source?
Also, domes of around 30k are not uncommon, do we need higher values to get the most detailed lighting?

[^Lele^]

So, i ran a few tests.
The attached contact sheet have four different cameras of a test scene.

In each row, from left to right, we have the default 2048 resolution for the dome IS, then 8192, and then the difference (modulo of A-B), with measured max value (as absolute float number, not a percentage).
Right below, the respective sample rates, and their difference.
The scene was rendered with a 3-100, 0.005 N.T., and without max ray intensity active.

The first set of images used a 15k HDRI of swiss alps, with a blue sky and a sun disk.
The second used some other HDR of Alpine roads, again with a sun disk visible.
The third set used a synthetic hdr i made in nuke: it's 32k* 16k, and it contains sixteen rows of R,G,B and White 8px squares, each at 750 float, separated a few pixels from each other.
It totals 2048 different light-emitting points, which i think is more than enough to cover the number of light fixtures in any studio, and they are incredibly tiny compared to the input image.
This is designed purposefully to try and break the IS algorithm, and should produce the biggest difference of all between resolutions.
Then another three tests with standard interior, or multi-source HDRIs from various collections.
Notice all contact sheets show the hdri used at the top,with info on the max brightness and the image natural resolution, for easier reading.

You'll find the lot here.

As you can see, the biggest difference is with the sinthetic HDRI, and even then, it'd be hard to spot if one was shown the two images separately instead of as A/B, although there is a definite brightening at the higher resolution (half or a quarter of a stop, perhaps?).
Most importantly however, the shadow shapes didn't change in any of the tests, and for all tests the differences are all below the noise threshold.
8K IS resolutions all rendered a wee bit slower (up to 35% in some cases), without really showing any improvement worth noting.

TL;DR: there is no need to ever change the default values.

[kosso_olli]

TL;DR: there is no need to ever change the default values.

So, hide the parameter from the UI, then. It may confuse users.
Anyway, thanks for the tests!

[^Lele^]

So, hide the parameter from the UI, then. It may confuse users.
Anyway, thanks for the tests!

It'd likely also enrage a few, if history's any teacher.

[Ihno]

Is there any point in converting 360° HDRI's to tiled exr's?
The Preparing Light phase seems to take just as long (is that the down scaling for importance sampling?).
Will the mipmaps be used for the reflections then? I did not see it using less RAM... but maybe my scene just needed the full 16k.