Dear people,

currently we are happy users of V-Ray CPU and usually render on our workstations and using online farms for animations. For some upcoming projects we are looking to expand our on premise solutions and are looking for a reasonable trade-off between price and performance. If it was easy and reasonable to solve, we would prefer to continue using V-Ray. Maybe I should mentioned that most of our renders are in the "indoor architecture" domain.

Our choices are :
V-Ray CPU (GPU has been tested with no success: many errors, some missing crucial features)
- this is our benchmark.

Octane
- looks good
- previous experience was that it crashed to often

Redshift
- Haven't been able to get it to look the way we want for indoor architecture.
- Seems stable enough, have delivered "motion graphics" projects with it before with success.

We are also eying 3delight, but it seems to be missing many features and in the end it is still a CPU-renderer where we might not use it's most prominent features like fast volumes and displacements.

The solutions that have been suggested are either to go with a rack of Dell Blade Epyc single socket servers or to get 1 or 2, 8 GPU boxes and render using Octane.

At first it seems like a no-brainer to go Octane in 2 boxes with 8-16 RTX A5000: Simple management and licensing solved. To match this rendering power, we would need about 20 Epyc CPU blades with a cost easily 4X that of the Octane solution.

So, this is all speculation and calculations. Does anyone like to chime in on their experiences?

The type of rendering you do should dictate the engine decision.
If you are doing Arch Vis internals, you cannot beat Vray, it's light cache stuff, and just general usage
for those types of shots is well proven and hard/impossible to beat.

For blades, it's usually better to not put your eggs into a few giant baskets, but spread it out more.
Any blade goes down, and that's a chunk of render power gone. So maybe look to not have 4 x beasts, rather
8 x lower spec for the same cash.

Vray will eat all your arch vis interiors, but will also scale to do all manner of VFX related renders too.
It is the primary renderer for us at Method Studios on everything from characters, to environments, to FX.

I think you are better sticking with Vray seeing as you are doing bread and butter stuff, but I'd point out
that unreal for arch vis is gaining a lot of ground, so it might pay to take a look into that arena, and start
devving a pipe for doing them in unreal on a decent GPU box.

L

Thanks for the input. Vray is indeed out default, but I must say that investing 50k instead of 200k makes it worthwhile to consider the GPU route. Hmmm, I guess we’ll have to do more practial evaluations on typical scenes before we move forward. Still quite interested in any other experiences with using Octane for “archviz” in contrast to vray or corona.
Oh, and I should mention that it is not really arch viz we are doing, but it is most similar in rendering needs. Real places, humans etc

Vray GPU - what are those errors and what are you missing exactly?

In short we did some tests and first and foremost the camera we needed for that particular test was not supported (fisheye). For normal camera shots It didn't look very close to the CPU images either. Like a completely different renderer actually... and finally it crashed when we turned on Optix/RTX. Seemed bad enough to not bother any more. But things change so maybe I should take another look