Realtime rendering versus offline rendering

[Robert Everhardt]

i saw these amazing realtime videos.
And i was just wondering how long would take something like this in vray rendertime, too complete such an rendered animation in vray?
i dont know the specifics,hardware/workflow but if an realtime renderer can do it as well as ofline renderer, saves a lot of time of rendering just static frames.
which sometimes are grainy or have faults in them, so you have too rerender, the gi calculation,calculate the precalculations, calculate the antialiasing , render the whole project again.
example if a client doesnt like something or want last minutes changes, it does makes sense now too do it rather in realtime versus offline rendering.
if i was making such animations in vray, i would have too send it to a renderfarm. only too see there where issues in some frames and resubmit the job.

offcourse vray is great, but too get good or decent results, it will take a while too go back an forth rendering/fixing static images.

so i still think, render too texture in 3dsmax could be better. which ultimately speed up the process.
off skipping the gi calculations, for static animations
or maybe even a faster algoritme for AA
and glossy/reflections?
https://www.youtube.com/watch?v=jaA2ZYhs-7I
https://www.youtube.com/watch?v=S2eXK025uC4


here is my example of texture baking in 3dsmax, i think my computer was busy rendering in 3dsmax too texture with vray for 24 hours
with irradiance mapping at medium 60/60 and light tracer at 1000, qmc off
and it still has errors
and because you cant use ambient occlusion in the render too texture

https://sketchfab.com/models/689eacd...b5dabeba612174

[Ihno]

From my experience it still takes quite a bit more effort to get something awesome out of a realtime engine than out of an offline renderer. This is mostly due to the fact that every poly and every pixel counts. You can't just throw a grass preset from forest pack onto a terrain and expect that to work. Its always alot of hassle with all tricks the realtime world has to offer.
Alot of faking is involved too. Like AO and normal maps.
Plus you're quite limited in terms of animation, especially when it comes to simulation.

No sss, no real translucency, very limited displacement, no render elements and at the moment no realtime gi on moving objects, no raytraced reflections and so on and so on.
Its right, they're coming closer, but they have still a long way to go.

The video you posted was rendered on a computer with four ultra high-end GPU's which is worth $60,000. It uses an experimental render engine that is still not production ready and not available to the public (as far as I know). They're just starting with that kind of thing and its far from being ready for us.

And don't think you'll work in realtime in current engines, even though the result does. You have to bake the light too and unlike vray there is no region render. You'll have to render at low resolution (lightmap resolution not actual pixels of the image) which results in blurry shadows and alot of artifacts which are making it hard to judge the final result. You'll have to render multible times and fix (much more) issues too.

I live in both worlds for more than 10 years now and for me working with vray is by far more responsive and straight forward.

There are some exceptions though, VR is the biggest one. Offline rendering just sucks for VR.

In the end its still working hours vs rendertime.
Rendertime is cheap compared to working hours.
When you start making at least one but often 2-3 lowpoly versions for every asset and uv all of them, bake maps for them and for characters even rig the same thing 5 times you'll know what I'm talking about <- thats not the case for small scenes of course but the buget is still very limited if you want to reach top-notch quality. Also, the realtime ready stocks are far behind of what we are used to get for offline rendering. Its very hard to find good optimized assets.

here is my example of texture baking in 3dsmax, i think my computer was busy rendering in 3dsmax too texture with vray for 24 hours
with irradiance mapping at medium 60/60 and light tracer at 1000, qmc off
and it still has errors
and because you cant use ambient occlusion in the render too texture

https://sketchfab.com/models/689eacd...b5dabeba612174

Vray for max isn't quite otimized for doing that kind of job. I wasted quite alot of time trying to get that to work good. You'll get faster and better results using the native baking engines of unreal or unity. Or, probably vray for unreal of course Didn't tryed it yet, though.
Take everything with a grain of salt since I dont know all realtime engines. For example, I have very small experience with unreal. But I think most of what I said applies for all of them.

[Robert Everhardt]

yes but do you need all that stuff if you can work arround it, find tricks
that makes things faster and still have the less or more the same result?
anyway here is also a thread about this topic

[Ihno]

Well, try it. We'll see
I dont expect wonders.

[AC5L4T3R]

At my old company, they've really ramped up the "interactive" department in the last year. They're all very talented and what they've created looks really great. It's a huge plus for clients to be able to spin a car around, press a few buttons and change their wheels/paint colour in an instant. Instant feedback, instant gratification. However, the quality is still a long long way off what the stills department are producing. I did Games Design in both College and University - 5 years of learning how to make games - but all I wanted to do was render cars with Vray and 3ds Max. Baking lights, unwrapping and all that rubbish was so tedious, I hated it and still do now.

[^Lele^]

Baking lights, unwrapping and all that rubbish was so tedious, I hated it and still do now.

That makes two of us.
And impressively, it still hasn't changed one bit in the annoying parts, has it?

[Ihno]

Yes, not much changed unfortunately.
I think due to 3d painting apps like substance it's a little less pain, though.

That Lumion thing highly profits from it's library it seems. As soon as you have to get custom assets in there you gonna be in the same situation.

On a side note: Since vray for unreal supports vrayscenes, does it support procedural's in object space mode? And the Triplanar map? That'd be pretty awesome!

[s.balabanov]

Procedurals are still costly for real-time and thus not supported on import. The triplanar however is but it is limited to where you can use it. We’d love to get feedback on this, so please give it a try once you have the time.

Best regards,
Simeon

[kosso_olli]

That makes two of us.
And impressively, it still hasn't changed one bit in the annoying parts, has it?

Still waiting for the all-purpose-one-click-unwrap-solution. Nobody likes that task...

[AC5L4T3R]

Still waiting for the all-purpose-one-click-unwrap-solution. Nobody likes that task...

Houdini is getting there, the game devs tools are really great.

I just find Unreal so hard to work with because the UI looks like it was designed by a 15 year old for 15 year olds.

[^Lele^]

Still waiting for the all-purpose-one-click-unwrap-solution. Nobody likes that task...

The issue isn't so much with the unwrapping per se, it's with the UV atlas use and consequent texture sizes.
We'd need pTex, or UDIMs, sure, but then we'd also need engines able to load more than 2GB of lightmap texture, f.e., as the texture sizes needed to avoid the UV bleeding would skyrocket.
Something like ID's megatexture (64k squared? can't recall.), or analogous intelligent Sparse Virtual Texture streaming approach.

If to the problem you add the fact that in VR people will do whatever they'll please (and of course get oh-so-close to that tiny object in the corner of your parking lot), you'll soon realise how baked solutions (read: textures, as opposed to per-pixel sampling) will fail sooner rather than later in most practical scenarios.

It'd be cool to try out, if and when it'll be ready for it, Lavina with a headset. The splotching by early denoising may or may not be unpleasant in VR, i don't know yet.
What's true is that no unwrapping nor baking would have to take place, which would sidestep both the issues above.

[Micha]

I have seen this thread here after I wrote my post here:
https://forums.chaosgroup.com/forum/...-mode-for-vray

I suppose so my request goe's in the same direction. Interesting maybe that Enscape allow real time rendering without that each polygon or pixel counts. An other engine like Enscape is Eevee. I don't tested Eevee yet, only Enscape. My experience from Enscape is, for conceptional design phases it is a great tool. For example I was able to render an exhibition design project - scene setup + 3 min fly through at 60fps/1080p + 30 high res interiors shots within 2 days of work. So, the advantage of offline render vs real time render depends on the project needs. My hope is that Vray could allow to work in a kind of game real time mode too.

[AlexP]

Most people ignore the amount of prep work required for good realtime rendering. Its all very well saying the frame takes 5 seconds to render - what about the unwrapping, poly optimizing, light building, managing GPU memory, performance, etc.
Its a balance, and depends on the need of the project. If you need a still only - offline rendering is still king because you can literally throw stuff together and hit render. Don't even stray from Vray default settings these days.
We're using Unreal quite a bit now, its amazing, but its certainly not quicker than producing stuff in Vray - unless you only look at the frame render time. (Considering a finished image, not development stuff)