DMC sampler and subdivisions

[Guest]

I have been reading this thread http://www.chaosgroup.com/forums/vbu...ht=dmc+sampler and this link http://www.interstation3d.com/tutori...yfing_dmc.html and i am torn between two ways of rendering.

Within the thread there is....

[QUOTE=Morbid Angel;520918]
DMC sampler is quite complex and hard to understand in all what goes on under the hood. Without going too much into detail, you may explore these settings:

Set all your lights to 128 subdivs
Set all your shaders to 128 subdivs (in glossy reflection)
Set your DMC sampler to 2/16
Set your DMC sampler settings to:
Adaptive amount 0.7
Noise 0.01
Global subdivs 2.0

/QUOTE]

My understanding is by using high global subdivisions, the DMC sampler must be lower in order to work out how many samples are needed for each pixel. If the DMC sampler is higher than the global subdivisions then it will result in not much adaptation and will lead to longer render times.

And then there is this method....

I have been trying out the method suggested by Toni Bratincevic @ http://www.interstation3d.com/tutori...yfing_dmc.html.
This method works great but takes a bit to get your head around. The first time I tried it I wasn't convienced and I didn't completely explore the concept but the second time around (about a month later) I was sold.

Basically you increase you're Adaptive Dmc Image Sampler Max value to 50. When you do this all the Subdivs on material and lights are divided by this number (50 in this case) when sampling, which will leave you with a number decimal number less than 1 which will be rounded up to 1. This means only 1 subdivs sample will be taken per eye ray. After that the Adaptive Dmc Sampler will kick in and start trying to reduce the noise until it hits the the Clr Threshold value. The problem with this is it takes along time.

So to counter act the High Adaptive Dmc Image Sampler Max value, we need to increase the subdivs on the Materials and Lights. Rember these value will be divided by the max number (50 in this case). So to get 8 subdivs sampled we need to input 400subdivs.
400 / 50 = 8
(This can be very strange to understand at first)
You could do this manually but I started to find this strange and annoying. So I decided to just globally multiply all subdivs by 50. By doing this I could still type in subdiv values that seamed familiar. (ie. 8,16,24,32)
Obviously mupltipltying subdivs globally will affect the Irmap HSph subdivs, so to counter act this I set this value to 2 (2x 50 = 100subdivs).

[ATTACH=CONFIG]6825[/ATTACH]

Here they are talking about the same situation but instead the DMC is much much higher and so are the subdivisions. Which is the better way of doing things?

[Neilg]

Those settings look way too high.

Personally, I drop the noise threshold to .008(dropping to .005 if needed) and up the material subdivisions based on how much noise is coming through in the material.
Same with the lights. if they're noisy, I add more subdivisions. I very rarely go above 32 on a material (usually 16) and the sampler itself tends to max out at 2,8.

I dont think anyone actually knows the best way of doing things. It depends on too many things.

[GLASS-CANVAS]

If it's of any interest I've always done things Toni's way (well, for the last 4 years at least), and I've always liked the simplicity, speed and quality of output you get when using it.

[Guest]

Those settings look way too high.

Personally, I drop the noise threshold to .008(dropping to .005 if needed) and up the material subdivisions based on how much noise is coming through in the material.
Same with the lights. if they're noisy, I add more subdivisions. I very rarely go above 32 on a material (usually 16) and the sampler itself tends to max out at 2,8.

I dont think anyone actually knows the best way of doing things. It depends on too many things.

Yes that is the more conventional most used method no doubt. Also leaving the min max at default 1/4 and increasing the max to anything up to 50 until the result is good enough.

If it's of any interest I've always done things Toni's way (well, for the last 4 years at least), and I've always liked the simplicity, speed and quality of output you get when using it.

Whats the difference between Toni's way and Morbids? I have just been playing around and it seems Morbids way is much faster than Toni's but that may be because of what objects I have in my scene.

[Guest]

Can any one shed some light on this?

[Morbid Angel]

I have quickly read over the tutorial and it seems very reasonable in terms of how it explains the DMC sampler. But what you have to take into account that while for some this may be just fine, in rendering stills and things like that, I have derived this method from a production environment. This is by no means saying that the explanation in the tutorials are not applicable, the method I have outlined is a basis for a certain work flow. In our productions the grain requirements are extreme. That is we must provide absolutely grain free render sequences. Combinations of lights and glossy reflections tend to vary between the DMC sampler settings. By giving DMC sampler a lot of samples to work with ahead of actual sampling you are speeding up the process of the sampling it self.

[PIXELBOX_SRO]

i agree with what Morbid says.
you can easily try this out like this:
render the scene with ALL the main passes using Dmitrys settings and compare with the other set of settings
youll notice that some of the passes will be very grainy but Dmitrys will be clean.

all in all...there is just so mamy combinations to achieve desired results in vray that its hard to set exact path.

Although i have to agree the DMC + noise thing could have been explained better in the online docs....especially now with new vray.
lets see what the upcoming SP brings

[Morbid Angel]

Well, I think also when you are talking about achieving a clear result, if you spend the time testing between adaptive subdivision, adaptive dmc or fixed, its safe to say that if you have a full frame glossy reflection, each of the samplers will achieve a clear result in approximately the same time, no matter how adaptive they are. I have found this to be true when I was trying to understand the correlation between adaptive subdivision and adaptive dmc. Ultimately DMC just lets you be more picky and precise with how many samples go where and so on, but all in all a clean render = long render time

[Guest]

I have never thought about working this way until now. I have been very traditional by setting the image sampler to min 1 and then adjusting the max and the colour thresh until I get a good result. Usually min 1, max 12 and a threshold of 0.005. I then adjust the material and light subdivisions manually until I am happy with the noise levels. Or if I am feeling lazy set all my subdivisions to around 32. Then use LC + IR.

In my line of work a little noise wont matter, its more about speed. I don't have the time on projects to be endlessly tweaking subdivisions and values, as we have a very fast turn around. I am working on finding a suitable universal setting that works for all, if not with minimal tweaking. Our render times are usually 1.5 hours for an interior scene using a dual Xeon and 8GB RAM. smaller scenes of product shots tend to take about 20 minutes. Resolution is usually 1600 x 1200 for quick renders that are needed yesterday and 2400 x 1800 for normal. I often wonder if these times can be improved by using different render settings. We don't do any compositing, just render out JPEG's that get slapped into a really bad Powerpoint by our sales team.

Animations seem to become a problem with the above settings the frame times are too long, I reduce the settings where I can but still some render times are 45 minutes to an hour at full HD. This may seem like a short time but for us this is too long.

It would be great to be able to leave all light and material subdivisions at 8 and then tweak the global settings to get a good result but I am becoming more and more confused by all of this DMC stuff. I too look forward to an updated explanation of the new Vray that perhaps is a little less overwhelming. I'm no mathematical expert by any means, I get lost when some of the experts talk about equations and how everything in calculated to actually render a scene, because of this I easily become confused.

[instinct]

Without having taken an in-depth look at the whole thing, the link above does not really de-mystify in my eyes. The information seems rather odd. Especially regarding Adaptive Subdivision sampling
to be weaker at preserving small detail. Give it a go. It is usually better for line rendering, and the example posted with the teapot is plain off..i can not recreate a rendering that bad at all using adaptive subdivision, not even when setting it to undersample with the default values.

Will have to dive in a little deeper after vacation but something seems odd on first look

Regards,
Thorsten

[m_hinks]

http://www.thegnomonworkshop.com/sto...torials/cni03/

Best money I ever spent! The theory in this is the basis for our every day scene setups. Just let the AA do all the work!

[greysheep5]

anyone still using the "universal" workflow with dmc sampler and minimum/maximum settings of 1/100, controlling the quality/noise simply over the noise trshld?

works quite well for us all the time i have to say, and not too slow. but maybe that´s because we don´t have ultra realistically lit indoor scenes with massive glossy effects all over most of the time...

cheers,

christoph.

[joconnell]

There's a lot of people asking about that now, there's a huge thread started by bertrand (bbb3viz) about the subject and how it fits with vray 2. The universal method always guaranteed a predictable way of getting clean results but not always the best time. With vray 2 it might not be able to get perfectly clean results by the same method, you might have to do a bit of chasing instead to find out what's causing the noise. On the up side, you'll get far quicker render times.

[greysheep5]

well, as i said, for us it keeps working perfect with the universal workflow... but maybe that´s because most of the time we don´t have hardcore photoreal setups, instead the occasional fill-light and stuff to keep it good lit and not too noisy.

and we use lightcache+irradiance map most of the time, no brute force.

cheers,

christoph.

[Morbid Angel]

universal work flow has changed with new DMC sampler, remember that originally it was designed around rQMC, and is now not quite the same. As I said before in my previous post, universal settings might be fine for rendering a still if you can have DR work on the frame for a number of hours.
The settings I have posted are oriented towards rendering thousands of frames at large resolutions, at which obviously you cant DR and yet you need to render with in a reasonable time frames.