Light multipliers too high, other bad oldschool habit & co.

[vypravci hubicka]

Hello,
Just started to build up new template scene for my archviz practise, based on max 2020 vray next, totally from scratch to avoid as many bad habits ( I probably carry with me since I started vray).
One question that came up to me was while setting up the night scene, I realized I am totally out of standart amount of light in the scene - noticed on the camera F-stop and time numbers ( something that should be say f-2 and 1sec for night, was aboout 1/60sec f-
This also leads to multiplier numbers for lights like 20,60,100 for omni and spot lights for lighting up interiors, streetlamps etc. and same or even higher (hundreds) for vray lights. Extreme example which stroke me in the eyes was when I tried to use lm units and tried to create small LED source, being curious if it will be plus minus realistic with my bicycle light. Created a vray plane light 5x5cm, multiplier 1600lm (which is my bicycle light) and the result was literally nothing. I had to use multiplier 160000lm to achieve about the same effect as real. Obviously this is down to HDRI setup ( I use the "old" Guthrie HDRI´s for night setup, which by multipliers set to 1,0 both for overall and render multiplier was too high and therefore the whole scene is "scaled up" light-wise - am I right?

Now the question: Is this something that is killing the correct computation of vray ( i.e. slow rendertimes, unpredictible noise, strange reflection behavior etc) and should be avoided generally?
And generally, is there any source/tutorial/thread generally describing how to approach archviz setup in 2020 to get good results within reasonable render times and quality? I remember Corona guys had a "bad habits" article pointing out some issues. I tried to search but did not find any condensed info sofar, however I have found some very usefull info from users like LeLe for instance.

There are actually two main things that make me nervous and force me to ask/search these things -

1. We run both Corona and Vray in the studio, where me and my team being vrayists and others are coronists. So obvious bitter jokes and discussions are daily bread. Unlike "them" I stayed with vray through its hard times ( which was about 2.5 and early 3 versions I would say) and also have done couple projects in Corona so I have quite recent experience and can compare ( unlike my colleagues, who damnated vray for eternity . There is a ton of reasons for me to stay in vray - from bucket mode over 2d displace, vray pattern and many these features - to properly working DR. However, there is one thing I have to admit - the materials, their reflections etc. seem to be more realistic in corona, leaving vray images being either slightly plastic, or requiring more pimping to get something as basic as plain architectural glass.

2. I am not convinced about our rendertimes. Although bulletproof reliable, I am not sure if full res exterior scene ( i.e. 5000x3500px, buildings, lots of proxies in cars, people etc, about ten FPacks etc .etc.) should be about 3-8 hours on a single machine or say 20min to 1hour on DR on my 13PC slave farm.
Vray setup mostly BF/LC 1/6 Bucket image sampler, 0.01 nosie threadshold, MSR and other setup left default. Of course, this setup creates the samplerate pass red as hell, but I was never able to get nearly as fast results with "proper" high subdivs ( like 1-24 default, or 1-100 for the socalled universal scenario). HW- wise, I ve got mix of quite recent AMD´s and some older Intel machines, everything on SSD and 48 and 64gb ram.


Sorry for a long post and thanks for any thoughs / directions on these issues.

[^Lele^]

If i got you right, yes, exposure and light intensities work in conjunction with each other (raise one, lower the other to stay steady).
Meaning that if you adjusted exposure in the 1600lm scene before hand low enough, it could indeed have made the light invisible.
Roughly speaking, V-Ray works with what is visible (the sampling adaptation works on the samples as they are after color mapping, which includes camera exposure.), so as long as the light isn't visibly shooting lasers, and you're not at either end of the floating point precision scale (very very very low or high), it shouldn't matter at all.

re: speculars. The models are nigh identical (both sport GGX, V-Ray allows for the full latitude of the GTR model's domain, however.), but i'm curious to see where you can pinpoint differences, personally. vlado and the devs may know of any essential difference between the two in code, however.

re: render times.
A wholly red SR is bad as It can mean one of two things:
a) you set up v-ray so that it reaches max AA subdivs before hitting the noise threshold, leading to an uneven noisy image.
b) you are dragging along some old scenario with local sampling active. You're then likely oversampling the image, and then getting into insane rendertimes.

The 1-100 approach has been entirely changed as of V-Ray 3.3, and is now the go-to method.
It's the one approach which will ensure sampling will stop as soon as it's good enough (i.e. the noise threshold has been met), instead of being wasted by overzealous settings, like in the old days.
To use it, on the scene where it's rendering too long, simply reset the render settings (switch to scanline and back), and without modifying *anything* render again. See if and how the SR changed, along with the rendertimes.

[vypravci hubicka]

thanks Lele,

Roughly speaking, V-Ray works with what is visible (the sampling adaptation works on the samples as they are after color mapping, which includes camera exposure.), so as long as the light isn't visibly shooting lasers, and you're not at either end of the floating point precision scale (very very very low or high), it shouldn't matter at all.
.

so if I got you right - it does not matter how much light energy is defined in the scene via multipliers. What you actually see in the camera is what matters? So if two scenes are brought to the same appearance - one via playing with light multipliers and the other via, lets say, shutter speed - this will not affect rendertimes/quality? (of course with an exeption of DOF and MB which depend on camera setup)

Also thanks for the info on speculars and render times, I will try to make some further tests/comparisons.