I’ve been employing alternative techniques for render setup using help from the guys on this forum, and have been learning from the workflow of, amongst others, Grant Warwick whose work is great. There are certainly some clever people on this forum.
Prior to the GW approach, we have been using solidrocks. Prior to that I was doing things manually, and bodging my way through it. Solidrocks has been very useful as we can quickly get ‘reasonable’ results without spending lots of testing time. The problem has always been that the results are only ever reasonable. The GW approach gives much more realism and control, but at the expense of render times. To a large extent, we should expect longer render times as we are getting better quality, but this hit is quite massive. For example, I left a couple of renders going over the weekend for a couple of 4k interiors and they were distributed so if all was stable, they used around 8-10 i7 PCs each. The renders took 12 hours and 8 hours to complete.
The results are great, but I can’t have renders taking that long.
Are there any archviz guys out there who are using the GW approach with high end interiors? Are you coping?
I have invested a lot of time with this and would hate to have to resort back to solidrocks.
On the GW lesson number 3, Grant discusses optimising and cleaning an architectural interior scene (supplied by Peter Guthrie I think). The thing is that this scene is soooo simple. There are a couple of walls, a floor and a ceiling and that’s about it. The types of scene we work with are luxury apartments - kitchens, bathrooms, living rooms with lots of furniture and fine details all over the place.
If you could get me a few “typical” scenes to vlado@chaosgroup.com, I will be able to give you more details on how to optimize them. We could profile the rendering to see which parts take the most time; sometimes there are mundane things like forgetting to turn off the 3ds Max bitmap pager, or filling the scene with ColorCorrection maps, or forgetting a Standard material somewhere, or maybe we didn’t optimize a particular use case… stuff like that is hard to figure out with looking at the actual scene.
I find reflection depth is the biggest culprit. you can slash a render time down to 1/4 or less with no noticeable visual difference if it’s been set up with 5 for every material.
Yes we’ve started using the GW approach, and Ive definitely seen a jump in render time but also a massive leap in quality.
Also agree that the scene used in Grants video was super simple, I cant think of any job Ive ever worked on that has been such a simple and empty space!
I guess it depends how much power you have in the render farm, but renders for us that take any longer than an hour at 4000px is not good enough.
There are def some inexplicable render time jumps using this workflow, but I put it down to lack of understanding more than anything else.
Also as cubiclegangster mentions, I’ve def found the same when reducing reflection depths.
I tend to have a gazillion colour correct maps in every scene do these really cause a big slowdown?
When I remember, I try to do this. Perhaps a quick run of the VMC on my scene is the answer - anything that looks black means I need to up the bounces.
I tend to work on Archviz interiors at about 4k almost exclusively and I’ve never been able to understand those comments that often get made on the forum along the lines of ‘a render should never take this amount of time, you’re doing something wrong’.
It’s like I must be using a completely different piece of software/hardware!
To me if you have a lot of VrayBlendmtl’s (these days), difficult GI because of the interior setting, a ton of complex textures, multiple light sources (the strongest of which is often coming in through windows), depth of field, displacement and possibly some hair/fur effects - these types of scenes are just going to take a long time to render. They just inherently require a greater amount of render power don’t they?
I agree the recent optimisation tutorials are helping massively in the understanding of sampling and streamlining times (as does the comments already made about limiting reflection depth)
but I still don’t imagine scenes of this type (high end complex interiors) flying by at render time. Either that or I’m doing something massively wrong somewhere along the line that I keep repeating!
I assume at present that every final render is going to be outsourced to a render farm and that cost forms part of the quoted fee. I quote interiors higher because of this.
Also, weirdly enough there seems to be a direct link between the expense of the real life materials specified and the expense of the render!
I am interested in any slowdown associated with the ColorCorrection Map though as I use them a lot (but then almost everyone must do). Would using the VrayHDRIloader for the same purpose be the preferable alternative?
negligible difference, dont worry about it. hdri loader helps when you’re running out of ram.
Some of the render times i’ve seen people talk about on this forum make my head spin. we work on high end interiors and over an hour for a 1080p frame is getting a bit high.
It’s bizarre - like I said I feel like I’m operating on a different planet from some of the users in terms of render speed. Mind if I ask what sort of times you’d average for a still on a high end interior (something like 4000x2800 px)?
I’m sure I’m being pretty careful with my set ups - one thing that did just spring to mind is that I’m usually a bit careless with overall polygon counts (always 3 iterations on turbosmooth at rendertime on most objects) but I thought that was more of an issue with RAM consumption rather than driving render times up incredibly. Don’t know why I assume that actually.
First off I am no expert, far from it. I too have had great results with GW tips and they have helped heaps with quality and render times. Once thing I must say is that even though he says to increase the samples by doubling them, just to see significant results, often you end up with too many samples in certain areas. We tend to forget to reduce the samples again.
Too many samples can have just a big impact as too few.
It’s just tight deadlines, issuing anything between 5 - 15 Hi res renders daily. Its easy to see why even the slightest increase in render time is a bit of a problem.
You’re right that certain scenes and settings just need more time - but it’s time we don’t have !
There’s a careful balance to be made between spending time optimising your scene and just using that same amount of time in render. For me it comes down to what I’m doing - if it’s animation then absolutely it’s worth spending a lot of time optimising to get frametimes down to keep render costs down for the farm. If it’s stills then for me an overnight render is perfectly acceptable across a few machines in DBR at 5k (talking about a complex interior with loads of lights and reflective surfaces etc.).
It just depends on what I’m doing and what kind of deadline I have.
In your example of producing 5-15 high-res images every day… if you’re pulling in that kind of huge workload then surely it would be best to just invest in another load of rendernodes and cut your rendertimes in half? In many cases, why spend 2 hours optimising a scene when you will only save 30 mins of rendertime? If you’re talking about animations it’s different of course, but for regular old day-to-day stills I find that more often than not I’d rather spend that time working on another scene/job and use that time rendering on my nodes instead.
It’s all about finding the right balance for the job.
Yeah, although I’m finding diminishing returns on more render hardware now to be honest. In order to cut times in half we’re talking serious investment, more machines come with their own problems as well, more licenses, more cooling, more space !!
Then its the fact that everybody needs the nodes available all day to be able render drafts and tests etc etc
Clients are expecting more and more drafts, sometimes up to 3 sets throughout the day, at at least half the final resolution.
Im definitely on the optimizing side of the fence, I like efficiency! I always spend time optimising, but then every minute I save doing that I end up filling with an additional effect or higher quality finish so it balances itself out in the end more or less.
Probably to need to re-explore render farm services… not sure they are cost effective for stills though.
In a particular scene that we profiled, 30% of the render time was spent inside the ColorCorrection texture (in that scene, almost every bitmap was passed through a CC map). Some more percent were spent in procedural Noise textures and such like. That’s why I said I need an actual scene to look at - it’s hard to figure this stuff in advance.
