I'd like to know some info about Redshift


#1

Apart from being faster, render volumes and having better light sources than native C4D, are there any other advantages that RS has to offer ?
For example:

  • Are there any Channels that the native Material Editor can’t handle ?
  • Does RS provide all procedural shaders of C4D like 3D-Spherical Gradients and Pavement ?
  • Does RS provide effects like Weathering, Pixel, Lumas, Backlight , Lens Distortion and Camera Shader ?
  • Is the native Material Node System less complete than RS’s ?

And a more technical one: Concerning only materials and not the lighting, what kind of material is impossible to make in native C4D renderer but not in RS ? (apart from volumes)


#2

I’ll leave the in-depth comparison to the people with more know-how; I only want to add a non-technical effect Redshift’s speed has:

It’s hard to overstate how much stronger your connection gets with your scene, once you have the super-fast visual feedback. It’s so much easier to playfully experiment with composition, fine-tune lights or materials… there’s just no going back.

Results are much better not only because it’s a technically very capable renderer, but mainly because you can iterate so much more in the same time (compared to eg. Physical Render).

On a sidenote: The Redshift-Team provides the best software support I have ever experienced. Sometimes it’s really needed too, because they release new versions in a fairly swift manner (like every 3-4 weeks) and they do introduce new bugs. But, as I said, it’s rarely a problem because of their fantastic support.


#3

I can only answer some of your questions,

Redshift does support most(maybe all?) procedural c4d shaders, via the “C4D shader node”. So you can use c4d shaders/effect. However, internally Redshift will render this to a texture map. Meaning that any worldspace noise or worldspace gradients don’t work. So weathering should work, but I doubt lumas would get predictable results.

I also have not yet found a way to make a 3d gradient at all in Redshift. I’m pretty sure it’s possible, but it’s not really as straightforward as native C4d.
And on top of that Redshift gradients (ramps) don’t loop.

I do love the custom AOV’s you can make in Redshift. … But I just now found out that are also possible in C4D’s node based materials since R21

Just like the previous poster said. The main advantage is the speed/quality.


#4

Does “render this to a texture map” mean that those noises cannot be animates ?


#5

It does animate. It will render it ever frame.
And specifically for noise, there is a “Maxon noise” node in Redshift that has all the fuctionality of c4d noise without the downsides of the “C4D shader” node. So 3d/world noise works correctly there.


#6

Does it support post effects like glow, lens flares, vignetting, color aberration, bokeh and highlights ?


#7

Post effects are of course supported Redshift has its own selection of effects but C4D’s native post effects like glow and lens flares will work too.

All of your questions could be easily answered by installing the Redshift demo and testing it yourself.


#8

My hardware rig does not allow me to have it installed yet. I had to make those questions because I couldn’t find any videos on youtube or references to them on the website.


#9

What keppn said about iterations is very true, and Redshift feedback is very quick most of the time,. But the same can be said for Corona, VRay, Octane, Arnold, these days. They all have interactive rendering to some degree, that allow for the careful fine tuning of materials and lights in near real time. Redshift does it well, but it can slow down real quickly with complex nodes, and it does have some weaknesses and problems compared to stronger engines like VRay that will not go away with their development cycles.

I’m speaking from my own experiences and they’re not limited, as growing numbers of clients are insisting that Redshift is part of their pipeline and that’s all there is to it. I would say that 7 of my last 10 jobs have been this way. At first it was a burden, but you get used to it like anything else.

I had a job with Apple, 3 months looking at a single scene trying to match a single reference photograph and an approved (Max) Vray version of the scene. I can tell you that for shit like Interiors, it’ll do an excellent job when pushed, especially with portals being able to push so much light into a scene. But seeing the ease with which the VRay guys matched the reference image with only a handful of lights and some beautiful material work, I found that Redshift required lots and lots finessing to get anywhere near (probably ‘50’ portal lights) the fidelity of a VRay shot - Like I said 3 MONTHS of pixel f**king for one shot… They paid, I delivered (and got numb to their daily changes after a while).

And more annoying still, touch a single dial ever so slightly in the wrong direction and your material often goes to shit, with a very finicky and sometimes difficult to dial in material system that often requires a multitude of material blended nodes to even get close to the quality of a Vray shader.

Redshift is phenomenal if you’re concepting, in the same way as Octane. You get fast results, nice renders and GPU rendering on a couple of RTX 3080’s feels lightening fast. Like any render engine, you can quickly choke a scene with too many fancy parameters. Redshift IMHO is piss poor with glass; does a lousy job at material illumination (glowy things lighting a scene), and struggles of course the same fate as all GPU renderers if you throw too much geometry at them. When you’re on a rig with 2 x 10gb cards, but also 256gb of RAM for the CPU side, big scenes, big environments require a CPU renderer, without a doubt.

So, overall, not knocking the software. It’s clearly gaining ground for it’s strengths, but it’s weaknesses are apparent and make it look very weak next to something like VRay.

Here’s 3 recent scenes that used Redshift exclusively.
https://www.behance.net/gallery/114908781/Stalker-1-Day-Project
https://www.behance.net/gallery/113040657/Air
And the opening segment here, rendering at 2 frames per minute: https://www.behance.net/gallery/113014017/Motion-Reel-2021


#10

@ThePriest
I like your feedback.

About the GPU/CPU you mention… What is the problem with large geometries and GPUs ? I thought geometry was a memory thing and GPUs are being favored for their hard-coded command set on trigonometric calculations and color. Do GPU renderers throw all geometry in the memory of the graphics card ? Even so, I thought that GPU accelerated renders like RS were hybrid… doesn’t that mean they try to minimize the computational burden by assigning some parts of the job to the CPU and others to the GPU ?


#11

I don’t know how it works programmatically! Perhaps it offsets geo to regular memory once it hits the GPU threshold. All I can say is that there’'s a noticeable dip in performance once you exceed a certain amount of geometry and start getting heavy handed with big scenes.


#12

Just skimming here so forgive me if its already been answered–but you can make 3d gradients/ramps, with RS although Its not particlularly obvious. It can be done with a rather complex math/vector node setup which are linked to 3d gizmo objects in the scene… A friend of ours set it up as an Xgroup node that we grab when we need to run a ramp through 3d space (linearly or spherically) between 2 points.

Also as of version 3 redshift has all th native c4d noise shaders included and no longer requires the internal bake to a uv texture like it used to which is great, although unlike native c4d, the noise paramaters in RS dont match the nosie parameters in c4d native. This is only an issue if you want your shader to match, say a displacer deforemer thats been applied to you object using the same noise. This may or may not have been fixed via using the newer native c4d nodes. Havent tested.

I have come to really enjoy redshift after using it exclusively for about a year and a half. As mentioned it can be fast when “iterating” during look dev, though you can also easily make it crawl to a standstill it if you throw the kitchen sink into your scenes. (for instance I had a file that took 12 hours to render 1 frame, though that can be chalked up to severe sloppiness on my part :slight_smile:) it does require some understanding of the render settings.

It can also be a little more challenging at the beginning to get a good look out of your work with RS until youve really learned it, compared to say Octane or Arnold which seem to give beautiful results with little effort. Theres a lot more settings to fiddle with in RS.


#13

No, Redshift is GPU only. Years ago there was talk about hybrid rendering but that never took off. It ended up being slower due to the disconnect between system memory and GPU memory. So Renders which support both CPU and GPU have split kernels that either render wholly on the GPU or on the CPU. Cycles does this but it’s inefficient as CPUs prefer smaller bucket sizes and GPUs prefer larger so you compromise the speed of one and end up with a marginal gain if at all.

Redshift requires that all geometry and textures are in the VRAM of the GPU to render at full speed. Once VRAM is exceed this leads to ‘out of core’ rendering where overflow geometry and textures are stored in system RAM. Obviously there is a significant speed penalty when you run out of VRAM as these data have to be transferred to and fro across the system bus.

If you want to render big scenes as quickly as possible then you must have GPUs with large amounts of VRAM. You can pool GPU memory with nvlink capable GPUs.

8GB GPUs are really only useful for pack and product shots especially on Windows which reserves a chunk of VRAM for itself leaving 2-3GB free for your scene. Linux gives you more to play with as it doesn’t steal VRAM. 11GB is OK but you really have to rely upon using proxies and repeats of textures for Archviz type stuff and small vignette type scene. The 3090 with 24GB is a good place to start as it also support nvlink.

For really really dense scenes with lots of subpoly displacment etc the only choice is CPU rendering which is why the high end still prefer CPU rendering with Renderman and Arnold.


#14

I work almost exclusively with Redshift now, thing that bugs me:

  • still lousy support for layered photoshop files (no support for layer sets). so your asset folder gets unnecessarily large
  • Weird support of grey scale images, questionable support of .tiff, best to avoid problem is to import every images as photoshop RGB so if you only needed 16bit grey scale your asset folder again grows unnecessarily large
  • no support for shader stacking, some things are done more efficiently out of the shader graph.
  • dome lights Environment are not fully controllable, excluded objects are still affected either through refraction or reflections at times it isn’t a problem but in other cases it is.

Renders beautiful glass over here, with gorgeous, fast caustics.