Cheatsheet: How to get a Cleaner image with SP3

[Pixelab]

Give the default BF+LC settings a chance, they may surprise you for speed and quality (you will see indirect shadows you would never have seen in SP2, for example), while not setting you back from IM+LC as much as they used to.

That's why I'm doing right now, but I'm often shaking my head reading setups here that takes 4+ hours to render a 4k, or discussing values of noise I can't even see (basically under 0,005 pre SP3) We all do care about nice images and thrive for noiseless and accurate solutions, but clients in archviz want fast delivery & last minutes modifications, and IM has always been the best solution so far to do so (and you're right about all the drawbacks)

My average for a final 4800 wide image is between 30m and 1h for a somehow clean image, I tweak the quality to achieve this rendertime, I hope I'll reach this number in BF+LC with an acceptable level of noise (my tolerance is quite high as you may have noticed

[glorybound]

So my 19-hour exterior is insane? I am obviously missing something or I need to do more things in post. I rendered this scene, pre 3.3 in 4 hours, but now I can't get a render unless I render it over night. I agree about client wants and needs; they can care less about a lot of things we do like fine noise you can't see unless you zoom in 1000%.

That's why I'm doing right now, but I'm often shaking my head reading setups here that takes 4+ hours to render a 4k, or discussing values of noise I can't even see (basically under 0,005 pre SP3) We all do care about nice images and thrive for noiseless and accurate solutions, but clients in archviz want fast delivery & last minutes modifications, and IM has always been the best solution so far to do so (and you're right about all the drawbacks)

My average for a final 4800 wide image is between 30m and 1h for a somehow clean image, I tweak the quality to achieve this rendertime, I hope I'll reach this number in BF+LC with an acceptable level of noise (my tolerance is quite high as you may have noticed

[^Lele^]

My average for a final 4800 wide image is between 30m and 1h for a somehow clean image, I tweak the quality to achieve this rendertime, I hope I'll reach this number in BF+LC with an acceptable level of noise (my tolerance is quite high as you may have noticed

When you render with BF as primary GI method, the initial estimate (i'm talking of progressive rendering, here) is a bit off.
This is normal, as that is the estimate when the sampler detects all the image to be outside of the Noise Threshold, and so it's giving you a readout for the max number of AA rays (say, 10000), times the full image (or thereabout) being sampled.
As rendering progresses, big chunks of the image (your flat walls) will clean up and not necessitate any more samples, and the estimate will give you the time left to get to 10000 AA rays, times the part of the image (say, a third now) being currently sampled.
As the image cleans up, the current rendertime and the estimated rendertime will tend to converge half-way through (in other words, you get renders in less than the initial estimate).

This isn't happening for IRMap as the GI is "baked" and clean from the first AA ray, so the estimate is more accurate right away.

[Pixelab]

So my 19-hour exterior is insane? I am obviously missing something or I need to do more things in post. I rendered this scene, pre 3.3 in 4 hours, but now I can't get a render unless I render it over night. I agree about client wants and needs; they can care less about a lot of things we do like fine noise you can't see unless you zoom in 1000%.

It always depends on the CPU power you use. We render with 4x E5 2620 Xeon CPU (48 buckets)

The good thing is that Moore's law & vlado are always pushing those number in the right direction

Honestly, when I look at the cheat sheet of Lele, my opinion (and I understand many will disagree) is that the first image is an acceptable draft, the second one is a final render, and the three others are overkill (noise & rendertime). My target Noise Thresold seems to be 0.01, but actually when I make a test on real case scenario, I've found that 0.05 is good enough for my taste and it renders under the hour (but there is a lot of red on some materials, I guess I'll have to up the AA max). Am I normal ?

[glorybound]

Yes, that is usually something that is never talked about. People give their render times, either high or low, and we base it on your own machines. I have a Quad Core Xeon ( think it is - 24 buckets) 2.8GHZ with 32 GB of RAM.

[eyepiz]

So my 19-hour exterior is insane? I am obviously missing something or I need to do more things in post. I rendered this scene, pre 3.3 in 4 hours, but now I can't get a render unless I render it over night. I agree about client wants and needs; they can care less about a lot of things we do like fine noise you can't see unless you zoom in 1000%.

Bobby, Was your pre 3.3 IRR or BF? from 4 to 19 hours seems insane to me.

[^Lele^]

So my 19-hour exterior is insane? I am obviously missing something or I need to do more things in post. I rendered this scene, pre 3.3 in 4 hours, but now I can't get a render unless I render it over night. I agree about client wants and needs; they can care less about a lot of things we do like fine noise you can't see unless you zoom in 1000%.

Bobby, we'd need a scene.
What you write isn't enough to ascertain anything at all.

[Streetwise]

I have to uncouple the Dome and raise its subdiv to get acceptable noise

Philippe, how do you uncouple the Dome? I've thought the same thing that it would be nice to tweak it's subdivs to get the level of quality that I'm looking for.

[Pixelab]

Philippe, how do you uncouple the Dome? I've thought the same thing that it would be nice to tweak it's subdivs to get the level of quality that I'm looking for.

As far as I understand it, you shouldn't if you're working in BF+LC. But if you want to, check "Use local subdivs" in Global DMC rollout. I find it useful in order to tweak IM+LC scenes to some extent (if you don't want to convert to BF+LC )

[glorybound]

I'll wrap it up and send a link. I had to wait for the render to complete

Bobby, we'd need a scene.
What you write isn't enough to ascertain anything at all.

[Streetwise]

As far as I understand it, you shouldn't if you're working in BF+LC. But if you want to, check "Use local subdivs" in Global DMC rollout. I find it useful in order to tweak IM+LC scenes to some extent (if you don't want to convert to BF+LC )

Thanks for that. Ever since 3.3, I've been using BF+LC and its done wonders in terms of eliminating a lot of blotchy interior walls. It may be slower, but I *really* love the results. Good to know on the Dome Subdivs though. I'll keep that in mind if I happen to go back to IR+LC.

[^Lele^]

Am I normal ?

Regarding what, exactly? ^^

Jokes to the side, yes and no.
Yes, because of three reasons:
1) noise PERCEPTION is subjective (its amount can be measured with non-debatable numbers), so you are entitled to like what it is you like, and let no one tell you any different.
2) The test renders are very even on purpose, and very far from real world scenarios (with complex geo and textures) because i felt those would better illustrate the sampling changes.
3) Because the numbers measuring noise in the images, simple and more noticeable the noise may be, show that the levels fall below what a standard 8 bit, sRGB display can show (1 point of gray in the 0-255 scale. that's 0.00391 in float money) right at image 2 (NT of 0.01, resulting noise of 0.0034, so below 0.0039 displayable by the monitor).

As a result, noise there is way more noticeable than if my surfaces had textures and complex shaders.
No, because of two reasons:
1) Noise perception is subjective, and there is someone's overruling yours: the client's. Ultimately, you'll have to deliver what is acceptable to them, and be prepared to meet the minimum requirements swiftly and without setup troubles.
2) While the sampler is less dependant than it ever was on lighting and shading variations across scenes, there may still be corner cases requiring a lower threshold, or more AA, than the defaults suggest, either due to some oddity in shader or lighting or geometric setups, or because there is just no other way around it to get it clean.
So consider these early tests more of a rule of thumb which works for the type of work you normally do, within the types of setups you normally use, than a general rule.
Your mileage may, and likely will, vary a bit.

[^Lele^]

As far as I understand it, you shouldn't if you're working in BF+LC. But if you want to, check "Use local subdivs" in Global DMC rollout. I find it useful in order to tweak IM+LC scenes to some extent (if you don't want to convert to BF+LC )

That's dangerous.
Doing so re-enables also all of the shaders' and lights' subdivs settings.
It's chance you got it the way you did. (i was left wondering how you managed to tweak just the dome... ).

Please make sure you know that doing that all but invalidates the new approach' benefits.

[glorybound]

This one was IRR + LC, however, previous attempts were BF + LC and were about as long. I didn't try Progressive on my exterior, yet.

Bobby, Was your pre 3.3 IRR or BF? from 4 to 19 hours seems insane to me.

[glorybound]

I'll wrap it up and send a link. I had to wait for the render to complete

I pm'd you the link