Understanding DMC Sampler

[joconnell]

On the way

[m_hinks]

On the way

What a tease!

[joconnell]

To be honest things have changed so much as other people have mentioned I wanted to let things settle down for a bit!

[peterguthrie]

looking forward to trying out the new "Subdivs mult" in the object properties thing!

[wyszolmirski]

I do have a feeling that most of what was written in this thread is totally irrelevant with newest V-Ray. Seriously, the best quality to speed settings we found are 'universal' + playing with min shading rate + max sample intensity.

Just do yourself a favor and don't waste time on playing with every shader / light / qmc subdivs, in most cases you will oversample the image and it will render longer than needed. Spend that time on the shading and lighting or go outside to do some sports.

Short tutorial how to do it:

Just turn on GI ( BF + LC ). In SP2 settings look pretty good by default. Than just use progressive (default) which is basically set-up like the old 'universal' settings. You can also switch to adaptive and crank up the max subdivs to something like 50 or 100 to get similar settings and be able to enjoy render buckets eating through your image. In both cases you control the noise amount by setting up the threshold. Once V-Ray reaches it, it will stop subdividing and sampling that pixel and you get exactly the amount of noise you want.

In both cases you don't need to touch sampling for lights / shaders etc... If you want to optimize it a bit, render out small region, write down the time and play with min shading rate, re-render, write down render time and repeat until you find the fastest value for that scene. That's very simplified but also very effective workflow, you don't really need to understand what's going under the hood.

Best,
Tomasz

[m_hinks]

To be honest things have changed so much as other people have mentioned I wanted to let things settle down for a bit!

Fair play, cant wait. I keen jumping between settings at the mo in the middle of a long frame interior animation trying to speed them up, with moderate success!

[JezUK]

*(deleted please ignore)

[^Lele^]

I do have a feeling that most of what was written in this thread is totally irrelevant with newest V-Ray.

Vlado has indeed been listening carefully, and changes have happened which are quite fundamental.
However, he managed to keep full backwards compatibility (want to slave away setting every single sample? sure. Want to drive the whole render from one place? sure, with more to come), and that required a very light touch, which may have made the changes somewhat less prominent to spot.

*Switch to progressive mode (the more cores you have, the merrier a case for you to do so. Progressive AA speeds are now IDENTICAL to the bucketed AA speeds, under the same noise threshold and subdivs conditions)
*Set your max AA to whatever you need to clean the thin/moblurred/DoFed Geometry and the high detail textures (for those, i suggest using a vraylight material as a tester)
*Drive final shading quality with the MSR at a global level, use the per-object subdivs multiplier to drive it up or down in a relative way to the rest of the shot.

There is hardly any need any more to touch anything else: returns are severely diminishing when one starts to tamper with shaders and lights, and the adaptive routines got a wee bit quicker too (so adaptivity isn't the killer it used to be when one was looking for the outright best possible render speed), so the case of progressive+MSR becomes even more appealing than it was.
The workflow could probably be better from a UI standpoint, but hey, work on that isn't finished yet.

As a side note: it's a very good idea, when a SP comes out, to make sure you are not using whatever you saved in the vray presets as "default".
This will make you miss out on the new, optimised defaults Vlado implemented, after he carefully profiled the various effect engines, and changed parts of them.
The suggestion is to save your "default" preset to a temp, delete it, and restart the max session, so to take a look at what's new.

[3LP]

*Switch to progressive mode (the more cores you have, the merrier a case for you to do so. Progressive AA speeds are now IDENTICAL to the bucketed AA speeds, under the same noise threshold and subdivs conditions)
*Set your max AA to whatever you need to clean the thin/moblurred/DoFed Geometry and the high detail textures (for those, i suggest using a vraylight material as a tester)
*Drive final shading quality with the MSR at a global level, use the per-object subdivs multiplier to drive it up or down in a relative way to the rest of the shot.

1*) Is that since SP2 already? I haven't tested yet but in SP1 it's definitly not the same time/noise between Bucket and progressive. I've read from Vlado himself that this will be only introduced in SP3. Maybe alreayd in the nightly?
2-3*) Haven't tested yet but it seems to be a great workflow, thanks

Stan

[^Lele^]

Err, you may actually be right about the progressive improvements being only in the nightlies.
I have to double check, i have the images saved in the VFB anyway.

[^Lele^]

Rendered with the latest stable release everyone has.



I'd say point 1) is also a decent suggestion (especially considering this is all rendered with defaults, GI turned on, GSD to 0, a higher MSR and different Noise threshold.), in light of the fact i only have 8 cores here.
Progressive is known to scale quite well and a bit better, if anything, than bucketed with the core count increasing.
A screenshot of my max UI with the settings i changed from the release's defaults.

[Vizioen]

I render a lot in that fashion, but I wonder about your noise threshold, that's quite an odd number.

[^Lele^]

it's 1.0/256, or basically telling VRay to sample more within the display device bounds.
With enough sampling, that's the highest noise threshold which will return a visibly clean image (provided, ofc, the render is used as is, without being stopped up, or brightened in any way).

[3LP]

Hey,
So I just opened a random scene.

Here are the results :

1) Universal bucket.


2) Progressive (1min)


3) Progressive (two times more render times than Universal bucket)


As you can see, bucket render is by far, faaaar, waaaay better than progressive. Even with twice more rendertime...
It might be because I don't have crazy settings as you do but I use the default settings everywhere.
I expect that with the new way Vray try to be more easy and trys to take away the need of optimising the scene, I would expect that if I use one or the other technique, it should al least be close to each other with defalut settings.
But it's not, even with twice more time, the progressive is really not good.
This is a simple example, I have this on pretty much all my scenes.
Progressive rendering has been there since the release of Vray 3, but I just haven't been able to use it in production, at all.

I might do something wrong (and I hope I do), and if so, I would be glad to be shown how progressive could indeed at least be closer to bucket rendering.

In case it's any useful, here are the two screens of my settings.



Stan

[^Lele^]

Of course progressive is slower: its noise threshold is set to 0, so it will always try and cast the max rays: make sure you match it to 0.01.
Regardless, bundle size is likely WAY too big: try setting it to something like 16, and try again.
With 128, you may notice your cpu occupancy isn't quite optimal, hence the slower render.
P.S.: with a region, bundle size has to be set correspondingly smaller, or the slack will be very big, and the region will render very slowly.
Always benchmark it on a full image if you're planning on a sequence, with final settings for bundle size (you'll soon get the hang of which is right for the number of cores/system you have), and if you're testing on a region, try and maintain the proportion with the image area (ie, if for 100% of the image i used 100 of bundle size, as a guideline for a region 10% of the image i could try and use a bundle size of 10.)

Example: