Can stuck bucket syndrome finally be fixed, please?

[kosso_olli]

That's looking great! What is the smallest bucket size after subdivision in these cases?

[^Lele^]

It's all radoslav.platikanov 's ingenuity and work.
Currently it splits down to 1 pixel, but we're still trying to find better ways about it, as in fact a 1px bucket requires another 8 pixels around it to be calculated for proper anti-aliasing.
Even then, we have so far found no scene going slower with this method than without, no matter the slack due to bucket borders.

One could imagine extreme corner cases where one pixel requires a googol samples, and so the LBS would still technically be possible.
So far this hasn't happened, but in any event we have ideas on how to treat that as well (but it's a wee bit more work.).

Here's another scene with great unbalance in sampling, screen-space wise.
Notice that the chandelier has been made spinning (hear my inner animator rawr...), and there are very bright lightMtl filaments inside glass at the tips.
Those are then getting a Bokeh and (curved) motion blur.
Also, notice that with smaller bucket sizes the mouse follow works just fine, so one's not limited to the previous sample of huge buckets.

Lastly, download the source file, don't use the gDrive viewer, to get it at original resolution.

[Paul Oblomov]

Interesting. Have you tried rendering phx foam or smoke ? They can be very tedious to clean.

[^Lele^]

Interesting. Have you tried rendering phx foam or smoke ? They can be very tedious to clean.

Whatever the frame content, the buckets will naturally clump where the higher sampling is needed.
So if it's the glinting foam bubbles needing the most work, you will find your buckets coming to the aid of those already at work on that part, should you have any free threads.

[Paul Oblomov]

And it is coming to gpu as well ?

[waaazoo]

Awesome. I'm looking forward to see this implemented in official builds!

[^Lele^]

Theoretically, yes.
In practice it needs to be evaluated as the benefits may not be apparent for the GPU engine, and/or it may not play well with the rest of the code.
We're not quite there yet.

[kosso_olli]

Looking forward to test this on the classic headlight or fast spinning wheel case we have so often.

[^Lele^]

What we can see from the three scenes of yours is that it works just fine.
On that note, ensure you do not use phong/blinn shaders, or perfect reflectors (i.e. gloss at 1.0), as we found plenty of those (from the looks of it, leftovers) and they are quite troublesome with your average setup (and are old and/or unphysical.).
Here's a simple script to sanity check a scene (if you want to run through a scene and make the changes, simply stand on the "alt+y" key combo).

[seandunderdale]

*Standing ovation* Very impressed guys! This is what the community has wanted for ages! Its fair to say most cases can be solved by better user settings, but sometimes its just nice to know the hardware we bought is working at its full potential all the way down to the last bucket.

[^Lele^]

It took some serious lateral thinking to get it done properly.
That it looks simple while at work, and that it wastes nothing of what was done previously, is testament to Rado's genius in tackling this.
All this to say we haven't been holding back on purpose, it's just very hard stuff to get right.
We're as glad as you guys (in fact, i've been giddy since i started playing with the build.), and have more ideas now (as it's often the case when some major obstacle is removed) to improve it to tackle even the nastiest cases.

[matthew999]

hats off to the vray devs, looks really cool will this beta feature be available in maya nightlys also?

[^Lele^]

Yes, it's planned for implementation in every DCC and in standalone.

[olly]

Cue the announcement for V-Ray 6...

[^Lele^]

Ahahah, not for a while, yet!