Right, I’m going to read that a minimum of 1 time, but a maximum of 8 times and see if anything sticks!
That’s almost how it works. Read it until the “noise” in your head gets clear + read twice harder parts :)…
Anyway RockinAkin - That’s great explanation. As it goes for DMC calculator, are you sure it’s correct at the moment for min shading rate? It seems like it’s still adaptive for secondary samples and it’s more like minimum[maximum secondary samples]. I base this on the fact that if you have scene with all subdivs set to 2, adaptive 1/6, shading rate to 64, changing adaptive amount from 0.85 to 0.0 makes it render longer (in my case), and overall rays are more than double, that wouldn’t be the case if min and max secondary samples would be set to 64 regardless of adaptive amount.
What’s going to happen in your case is that you might be happy with the quality of your GI at 64 samples, but because vray isn’t, it’s sending back a noisy result to the AA sampler to finish off, and the aa sampler is working on it again (green sample rate) and putting up your render times. Raw GI and Raw light should be grain free in their elements with only the samples of the lights adjusted. It’s a bit counter intuitive as you might be happy with your times at the samples you’ve got and not want to touch them, but try upping them for the laugh and see if it gets better at the same render time with more samples, or potentially faster with the AA having to do less!
I will have a try, though my testing time is seriously running out here! Seriously though, I’ve never really upped values above 32 or 64, so to me it seems ‘obscene’ to raise them any higher. What kind of values do you tend to use (for lights like the vray sun, a coffer vray mesh light, a vray ies spot light) for instance? Likewise with BF GI - What kind of figure do you typically get to?
More often than not in the past, I have left all subdiv values at their default value (8 for materials and lights, 3 for the sun) and only upped them if I have seen major issues. This is a very new way of thinking for me.
I am persevering with this method of applying a grey material and sorting out the image sampler, and then applying the lights and GI and adjusting the subdivs of the lights and GI until their respective elements become clean. It works quite nicely and it is quite enjoyable. The results are rendered reasonably quickly too.
But - and its a big BUT - as soon as I switch back to my realistic materials, the render slows to a crawl. Not just a crawl - buckets don’t even appear! I have used VMC to return all the reflection subdivs to 8 (even 4). Could this also be counter intuitive? Should I be increasing these to 100, 200, 500…?
This is a different scene - nothing too radical. An interior with a timber floor, a bunch of ies spot lights. A vray sun and a few coffer mesh lights. Its frustrating to have got so far but stumble at this point.
For a dome light in my current scene which is around 1/12 I’m using a dome light and a large sphere light for a fill - 128 on both.
You’re right about the high values but there’s two things to factor in. One is that your values are being knocked down by your max aa, so me using 128 on my dome lights with a max of 8 aa automatically brings the 128 / 8 = 16 so it’s a less dramatic number, also you’re only giving vray the possibility of using that number of values if it needs it, there’s no guarantees that it actually will. With vray being adaptive, it’ll use the high subdivs if it needs it, but not bother if it doesn’t.
No luck my end I’m afraid. Switching back on my materials just stopped vray rendering any buckets at all, even after 10-15 minutes.
I tried changing my GI method from BF back to something I’m more familiar with, LC/IRmap. The GI calculated quite quickly, but again, no buckets rendered after 10+ minutes. I globally switched off all reflections and glossiness but still the same problem. Finally, I reset all my lights back to 8 subdivs and this seems to have a big impact - the rendering starts almost immediately. Obviously, now my rawlighting looks grainy and my rawGI looks blotchy (as you’d expect from irmap). Switching back to BF cleans up the rawGI with the normal render time hit I’d expect from using BF.
Given that you gurus are adamant that the step by step process of cleaning the rawlighting and rawGI elements is a great method for controlling detail etc in your renders, I can only assume that something I am doing with my lighting rig is different or not efficient for this way of working. I use quite a few vrayies lights as spots and normally sweep a line for a ceiling coffer light (converting the sweep object to a vray mesh light). I use a few plane lights (long thin ones) sometimes for other types of coffer/strip lights. I use the vray sun/sky system generally rather than HDRI lighting.
Puzzling, but I fear I may need to revert to solidrocks or my own ‘hybrid’ methods in order to get jobs rendered in time for deadlines. I will continue to test when I get chance.
EDIT:-
This is as far as I got with a project using the step-by-step approach. I turned the coffer mesh lights off to try and speed things up. I also dropped the samples of all the lights down too - earlier tests had some of them cranked up to 128 or more to get rid of noise in the rawlight element (particularly the coffer mesh lights and strip lights in the kitchen). You will see that the rawGI and rawlighting elements are far from clean, and the samplerate element still has quite a bit of red. I also globally switched off all glossies and removed any trace of reflection from the walls and ceiling in an attempt to speed things up. Subdivs on materials were globally dropped to 8 using VMC. This is an 800x800 pixel render and it took 32mins rendered across seven i7 machines with either 16GB or 32GB in them. The scene uses only around 2GB to render. We typically render stills at around 4k, so I would assume this will take several hours (overnight maybe) to render on our farm.
Settings are as follows:-
Image sampler min=1 max=8, clr thresh=0.01
GI LC=1000, BF=128
ceiling spots = 4subdiv
vraysun=4subdivs
kitchenstrip=8subdivs
Tricky u could try to avoid the vray mesh light and try to apply to the mesh vraylightmtl with direct light not checked. As in this thread:
You could get faster render times
Hope it will help
If you had clean gi with no major issues in the sample rate, then it might be that your material subdivs are now way too low and when you turn the override off, you’re getting crucified with the old settings. I’d finish this project off then start a new one with maybe four materials and two light sources and use that as a handy test bed.
Wish there was a thank you button on this forum, instead of bumping with a quote. But thank you very much joconnell!
It would be interesting to see how V-Ray 3.0 handles this. Do you think you can maybe get me the scene to play with it?
Best regards,
Vlado
Hmmm - I’ve confirmed this behavior on my end as well. I’ll look into it further and will update the formula to fix it.
Great Akin!
p.s. Thank you for the calculator, it’s very helpful.
No Problem! I just updated the calculator with an updated formula for Min Shading Rate. Let me know how it works on your end! (You may need to fully refresh the browser window and cache).
Works great. Cheers!
I will see what I can do when I get a moment.
I may be having a breakthrough moment here. Managed to drop down my light subdivs. The biggest culprit were vraymesh lights. Once I went back to my old-fashioned methods of using vraylight materials, those issues went away. I have also reduced the reflection depths of most/all materials to 1 or sometimes 2. Reflections have generally been sharpened up so I have been able to reduce the samples there too. A subtle timber bump texture on my floor, which added very little to the beauty pass, has also been ditched.
What times are you getting now on that 800X800 render with materials active?
I managed to get a full size (4000px) render done in around 1.5hrs using about 6x i7 machines. Its a simplified model with all the furniture removed. The elements look pretty clean - even the samplerate element looks ok. Still areas of red on edges and on areas of my timber floor, but not bad overall.
Please post your times when you do the final image. Using this method and also doing 4K images with 7 X i7s I’m getting around 3 to 7 hours for interiors and 1 to 4 hours for exteriors. Most of them pretty complex. I like how clean things come out, but just thinking back I use to get these times with just a single dual core pc. Now I’m getting the same times but with 7 X high power computers. Things do look better however, but the scale is not proportionate. Now I’m throwing 5000 % more power than about 10 years ago, but things only look 3000% better. Makes me think in the future how much power I will need to render an image in 5 or 10 years time, just to get marginally better images.