Understanding DMC Sampler

[snivlem]

I agree, I always see 1/32 as being usually the fastest. 1/8 always leads to hours worth of extra rendering time for myself regardless of material subdivs and noise threshold.

For instance on my project at the moment, 2/32 is over 100% faster than 1/4 1/8 or 1/12 and with cleaner noise. Regardless of material/light subdivs. Usually lowering them makes render slower.

For me, I have found that if I set the Noise threshold high .05 but have lights and glossies set to quite high subdivs of around 80-256, it's really quick and noise is acceptable when considering rendering speed and for the purpose of a still image. I think this is similar to Peter Guthries settings a few pages back.

[Morne]

This is really strange... are you sure no other settings have been messed with?

Yes I'm certain

Any thoughts?

[stef.thomas]

Wow, this thread is a bit of a revelation to me. I had always discounted DMC as being slow and gotten quite good at tweaking my LC and IR Map setting to get nice clean images. But having recently decided to increase my standard output to 5k from 3k render times had been getting too much. The last batch of images I just completed were pushing 10 hours per render. This particular interior scene has a lot of lights (dome, plane and IES) and glossy materials.

After converting my scene lighting and materials using this method I feel much more in control of render times again and can get those render times down to 2½ hours, with only marginally more noise! The process of going through the individual elements to identify causes of noise fits in well with the workflow I was already using to set my GI, so feels very natural.

I reckon I would have to lower the colour threshold back down a bit from 0.03 on the 2½ hour render to be happy with the output, but cutting the time by more than half would still be a great result. Overall I would say that the lighting solution is also a little brighter and better defined. Thanks for your input on this everyone

[pengo]

2RockinAkin
I get the same impressions/results Snivlem gots using high ratios between AA min/max values, like 1/32 or 1/64 respect 1/8 with proportioned subds in mats lights qmc subds and so on. Faster with high max values respect lower as u gently explained. I'd be very interested if u could tell more about your settings to understand where I fault. I use subpixel checked on (this is not suggested by the error list in vray messages window, so I underlined the fact is active) and clr threshold 0,01 in my test and noise threshold 0,01 too (i didn't respet the universal settings in this case that tells 0,005 or less for both).

Thanks and thanks in advance
Sorry for my english

[joconnell]

Snivlem and Giacomo:

It very much depends on your scenes - they could be very simple from a materials point of view or easy to clean up so the aa sampler is able to handle everything - in other words the materials or shadows aren't stressing the samplers at all so your aa can deal with all of it. You might have scenes where the universal method is totally able to handle everything, for scenes where the materials and lighting become less kind however, trying to clean everything up by AA methods alone might not give results as nice or it might never reach the level of cleanliness that's required.

There're no settings that will ever cover every based in vray so in some ways it's not really useful for users to share them, they'll work fine on their specific scene but not be optimal on others. One issue is that while the parameters have been explained, they haven't been explained in a friendly fashion so a lot of users don't have an approach to sampling for each scene. If you end up doing the same type of scene over and over again and have arrived at settings that work well for you then great, but say for example you do a totally different type of scene next time, maybe going from doing exteriors to interiors, or something with lots of clean materials to something with lots of texture and glossy reflections, then your sampling settings will likely not behave as well for you.

[Morne]

If I have a fairly simple exterior scene lit with an hdri or vraysun/sky and I have a number of stainless steel pipes in the form of a balustrade and everything in the sample rate element is pretty clean, except for some RED areas in the middle of my pipes, do I up my HDRI dome light samples to get rid of that? It takes much longer on those stainless steel elements. Glossy on 0.9 fresnel NOT ticked and subdivs of that material already on 128

[joconnell]

For really overbright stuff like that and especially if it's localised you've got two things to deal with:

1. The sampling of the material itself - is it noisy in the final picture?
2. The dynamic range - The image sampler will have some issues trying to smooth out the transition between extremely bright pixels to the darker ones beside it.

Could you pop up a crop of this section with some elements to show the sampling?

[Morne]

For really overbright stuff like that and especially if it's localised you've got two things to deal with:

1. The sampling of the material itself - is it noisy in the final picture?
2. The dynamic range - The image sampler will have some issues trying to smooth out the transition between extremely bright pixels to the darker ones beside it.

Could you pop up a crop of this section with some elements to show the sampling?

I can't post at the minute, but the final render is clean and noise free. It's more a case of dynamic range I think. Those spots VERY hot and probably reflecting the sun...
Would I then up my max on my AA or would I up my subdivs on my light?

[joconnell]

Something like that is a tricky one as it can be handled in three areas:

1. Better reflection samples (which will get more ray samples over the surface of your object so it'll see more samples along the edge of the light and smooth things out).
2. More light samples (which kind of does the same as above except rather than the reflections using more samples and "seeing" the light more finely, the light shines on your object more finely - more rays to smooth things again)
3. More AA. AA deals with edges / contrast and you've got a nasty situation where normal pixels of brightness 1.0 or less are meeting pixels of way over 1.0 so you've got really big values to transition across.

So looking at the three options I'd think the following:

1. This is the one for me to start with, if it's small areas of problems with the overall image looking fine then I'd up the reflection samples of this material.
2. Will solve this issue but it'll also up the samples for everything else hit by the light - vray should in theory not use the higher samples of the light but let's not give it the option of it if everything else in your passes is clean
3. This is very global unfortunately - it'll mess up all the rest of your materials sample values and throw off all the other things you're happy with. Per object AA settings were being looked at but they're not in vray 2.4

The other thing to think is that all of these issues are being caused by that dynamic range, so why not think of attacking that instead? Turning off the sun to reflections and speculars and making a new sun object that's only visible to reflections might give you the same look but with a more controllable highlight value. Vlado's reinhard burn trick might do it too to drop down some of the overbright bits? Lastly maybe just don't deal with the problem at all and clamp the render? If you're doing a load of post that relies on the overbright values where you're using blurs / glows that need the upper values, maybe clamp at a value that still gives them plenty of information - about 4 or 5 perhaps?

Those really bright highlights are tough ones - since they're so localised it's almost not worth trying to mess up all your settings just to accommodate them when they're the areas that get glowed and flared in post too.

[pengo]

joconnel, thanks

what I asked was to understand if the settings using not so high max value in AA like 8, as the mentioned user used, respect an high value like 32 or above, using at the same time an adeguate number of sudivisions in lights, materials, BF subds and so on, was faster in absolute as seemed to understand from RockinAkin message. And if so why I was getting the opposite effect, longer rendertimes. He used hdr map too as I use, even if I refer to an architectural interior.
But as u underlined and u perfectly explained depends always on the scene we are working onto. Thanks again for the continue replies to this thread. Really helpful

[joconnell]

Glad to help!

[pengo]

Always very helpful! Especially to a person like me, who always trusts that there is always a faster way to get a rendering with vray, even without loose quality. Thinking that I'm not the best optimizer in settings to get fast/resonable rendertimes, I always tend to believe in an absolute workflows to reduce rendertimes.. I'm an illuded person for sure..

[tricky]

This is a great thread. I'm pretty late to the scene, but a reasonable amount makes sense to my 'sometimes-slow' brain. However, as often happens in these kinds of threads, what starts off unclear gradually gets clearer with more explanation and example, but then other posts throw spanners in the works and my confusion wells up again!

I love the splitting up the sampling approach, but I'd like to confirm the process in 'baby-talk' so I can see if I have understood some/most of it.

Up until now, we typically use LC/IRmap for rendering. I am now using LC/BF - is this what most people are opting for, or BF/BF?

STEP 1: Work out the basic aa with default lights and a grey material - use the samplerate element to try and have all the scene blue with the lowest possible settings for min/max dmc sampler. Once min/max values are established, don't touch this again.

STEP 2: Switch on the scene lights and also render the rawlighting/rawGI elements - increase the samples on the scene lights until the rawlighting element look smooth and clean. Increase the subdivs in BF until the rawGI element looks smooth and clean. Once this is done, the samplerate element should also look mostly blue

STEP 3: Switch on materials and also render the rawreflection element - increase the material glossy samples until the rawreflection element looks clean

STEP 4: Render the finished image. Job done.

Steps 1, 2 and 3 can be performed at lower resolutions. However, step 4 (obviously) must be rendered at the full resolution, but we can use the same settings we have just worked out. Is this correct?

[tricky]

I'm working on a small interior at the moment and am using this to experiment with. Here are a few of the images:-



This took around 15 minutes when using DR across 6 high-spec machines (i7, 32GB). It is 1000x1000 pixels for my testing. I have applied a mid-grey material to everything.

Adaptive DMC image sampler Min=1, Max=16, Clr thresh=0.01
Primary GI = BF with 256 subdivs
Secondary = LC with 1000 subdivs

Does this seem about right? The elements look clean, but I am a little confused by the vraymesh lights I have running along the ceiling. In the RawLighting element they are showing as black. Is this normal?

[joconnell]

You are totally correct Richard.

Step 1: Yep - what you're looking for hear is using the right amount of AA to get your edge quality as you want it. One caveat here is that textures and bump kind of count as edge details - the AA sampler is looking for contrast between pixels values so typically that's at the edge of objects but it can be anywhere that there's large contrasts so when you turn on your materials, if you've got a map with loads of small fine detail in it such as wood grain, this will be something that the aa sampler will be looking at. In Morne's example above too, where you've got really bright values meeting dark values like a sunlight casting a really burnt out shape on a carpet, the aa sampler sees this contrast where dark meets bright as an edge too. I'd still go with step 1 as you are above though and try to resolve those specific issues in other ways.

Step 2: Bang on. With your lighting element you've got direct raytracing happening so that's light sampling only. Your sample rate may not be totally blue, but it'll be appropriate so that it's mainly blue but with red samples along your geo edges and lighely some green in your soft shadows.

Step 3: The raw reflection is important with one caveat. Vray will boost the reflection data in the raw ref pass so it's at 100% reflection so it's gaining up the values in this pass hugely. If you've got a material that's 5% reflective, the raw ref pass will take that 5% and boost it up 20 times. The thing is of course that in areas that are quite faint or don't add a lot to the final result, vray won't spend much time sampling these. If you take those quickly sampled results and boost them up 20 times they'll look very grainy, so you have too look at your raw reflections but also your main reflection pass to judge the quality of them as they are intended to be used in the final pass.

Step 4: Yeppers - Resolution is a bit of a tricky one since fine details will only be visible at your final resolution so even when you're looking at your edges in step 1, there might be fine detail that was too blurry for the sampler to notice so it's a little bit risky. Likewise if you've got a fine bump map then a lower resolution of this is easier to handle than a higher one so you're likely better off doing small regions of your scene at full res when you're tweaking material settings rather than lowering your res.