Redshift vs Octane in 2024?

[MJV]

16 hours ago, Mash said:

Can you upload the spheres scene, id like to take a look

I added the file to the post. 

[zeden]

On 12/2/2023 at 2:01 AM, MJV said:

I personally love the look of Octane's unbiased path tracer. It's the closest thing to shooting with a real camera as you can get, but as with even the most expensive real world cameras, its big weakness is noise. Scenes that use motion blur and camera dof will look beautiful and dreamily real, but scenes that should depict a fixed, razer sharp camera image cannot be rendered completely noiseless, even with prohibitively high sampling settings. The scene below used  16,000 samples and 3.5 minutes to render. More samples make no difference after a certain point, and so pixel peeping such a scene rendered in Octane is unlikely to be very satisfying. In such cases biased renderers such as Physical and Redshift will render much cleaner. 

 

OctaneWeaknessMV2.c4d 566.58 kB · 1 download

You don't use coherent ratio, no adaptive sampling, no denoiser and no multi GPU rendering. Gi clamp is too high for that scene. These are non optimizedrender settings. Sorry

Edited December 3, 2023 by zeden (see edit history)

[MJV]

1 hour ago, zeden said:

You don't use coherent ratio, no adaptive sampling, no denoiser and no multi GPU rendering. Gi clamp is too high for that scene. These are non optimizedrender settings. Sorry

 

This is a simple scene to show a simple concept. You know, or should know very well that AI is not a be all end all solution to noise artifacts. If the spheres were moons for example the denoiser would completely erase the craters along with the noise. The other options you mention don't do anything to make things better, but are ironically biased rendering techniques. So your solution is to cover up the artifacts with AI, throw more render power at it, or use biased rendering techniques to prove that an unbiased renderer doesn't have certain weaknesses compared to biased renderers. 

[MJV]

This is the same scene converted to Physical render which renders in 31 seconds. I don't use RS much but I expect it could do even better. 

 

OctaneWeaknessvsPhysicalMV1.c4d

[Mash]

Seems a bit extreme. You're using a light with a brightness of 3400 (your 1000 strength light material has no texture source, so it defaults to linear colour grey, so many of the samples are only doing a third of the work they could be doing) So 2/3 of the samples are wasted. Your material is mid grey, so half the brightness is lost off the first bounce then only half the brightness is received, so 75% of the samples do nothing. Multiply the above together and it means 90% of the samples arent doing anything useful.

 

Adaptive sampling is off so 90% of the samples are flying off into space to do nothing. We're down to 1% efficiency now? 16 light bounces are very wasteful on a scene where only 1-2 bounces can possibly have a meaningful impact. So we're down to... 0.1% efficiency in terms of render time.

 

Youre using geometry to emit light instead of a light. Octane can more efficiently render a scene if the bulk of the light comes from a light instead of polys.

 

When you have this many samples, you can 100% turn the denoiser on.

 

And finally, the scene is so utterly beyond overexposed, you're effectively cranking a noise with a brightness of 5-10 (0-255 scale) up to 100, which is going to exaggerate the noise you see. This is the worst case scenario of worse case scenarios.

 

As for the c4d physical engine version. You're using irradiance caching, this just slaps a bit of light on, blurs it together than adds it to the brightness. Effectively c4d is running a denoiser on the secondary light, something you're not letting the octane engine do. An apples to apples comparison would be setting physical to QMC + QMC with 16 bounces, but you can't because the slider tops out at 8 bounces.

[zeden]

On 12/3/2023 at 9:46 PM, MJV said:

 

This is a simple scene to show a simple concept. You know, or should know very well that AI is not a be all end all solution to noise artifacts. If the spheres were moons for example the denoiser would completely erase the craters along with the noise. The other options you mention don't do anything to make things better, but are ironically biased rendering techniques. So your solution is to cover up the artifacts with AI, throw more render power at it, or use biased rendering techniques to prove that an unbiased renderer doesn't have certain weaknesses compared to biased renderers. 

 

The denosier needs to be balanced against sampling for details like Mash describes above so accurately. These are state of the art render optimizations in the year 2023 which make Octane what it is. Octane is the master of high end algorithms combined with rendering shortcuts. The genius of Otoy was their balancing of all the tricks under the hood which made it the renderer so fast to kick off the age of GPU rendering. Throwing it away takes the strength of Octane away.

 

Maxon is implementing multiple of Octane's approaches into RS already and even more to come. For example RS RT's new experimental implementation of pixel based displacement mapping is basically the technique Octane uses for years so efficiently.

[MJV]

13 hours ago, zeden said:

 

The denosier needs to be balanced against sampling for details like Mash describes above so accurately. These are state of the art render optimizations in the year 2023 which make Octane what it is. Octane is the master of high end algorithms combined with rendering shortcuts. The genius of Otoy was their balancing of all the tricks under the hood which made it the renderer so fast to kick off the age of GPU rendering. Throwing it away takes the strength of Octane away.

 

Maxon is implementing multiple of Octane's approaches into RS already and even more to come. For example RS RT's new experimental implementation of pixel based displacement mapping is basically the technique Octane uses for years so efficiently.

 

You could write company press releases and promotional material because that sure sounded like it but I thought this was a discussion about Octane vs Redshift in 2024, which really boils down to a discussion of biased vs unbiased rendering, as things currently stand today. The thing I want others to know about Octane, if you are comparing it to Redshift or any other biased renderer is that, mostly because of the nature of what it is and how it works (and limitations of today's computers rendering power) an unbiased path tracer,  it has a very high noise floor compared to what can be achieved with biased rendering tools. Octane's big innovation is having a great denoiser. Without the denoiser it would be pretty hard to impossible to do almost anything in Octane.  And yet anyone who uses Octane knows very well it can be a tricky balancing act between preserving necessary detail vs reducing noise. This can be especially tricky if the detail in question is itself noise (as with a noised based bump map), or noise-like (small craters on the backside of a moon, metal flakes in car paint).  And it's a destructive process, that is, once you denoise it, you can't go back to the un-denoised state that contains the original information unless you deliberately save a copy of that as well.

 

Biased renderers have a bunch of different scene sampling methods and options that let the user dial in where and what and how rays are directed and how things are sampled (monte carlo, semi monte carlo, directed, targeted, etc.). The downside to this is, 1) it sometimes feels like too many choices and 2) a possible lack of photographic realism resulting from the different user settings and scene evaluations methods composited together into the final result. If you look at Maxon's old Physical renderer for example, different settings here and there can lead to vastly different results, but the final result may be best for product advertising or architecture or whatever because it looks good and clean and not because it is shaded photorealistically perfect. A biased renderer should probably also offer additional compositing options, since the render itself is a composite. 

 

I personally prefer using Octane because it looks most real to me compared to anything else I have used, and I simultaneously dislike the layered or composited look of biased renderers. That said, I wanted to acknowledge the weakness as well as the strengths of both. 

 

 

[Decade]

I'd agree with the general consensus that I prefer how Octane looks out of the box & I think it's faster overall.

The software & content bundle Octane comes with is pretty fantastic, in my opinion. You get a lot for your money especially if you work in a company where there are animators that would use Cascadeur, Environment Artists that would use Kitbash etc 

 

There are some advantages to Redshift though, that are more 'Pipeline' or 'Team' advantages

Redshift is more Neutral/ Plain in it's output, which can be an advantage for some types of work, especially where a lot of the look is done in comp.

The Viewport display of the materials is much better than Octane, which if you do a lot of previs/ playblasts is handy.

Everyone with a Cinema 4D licence can open the scenes & see all the materials correctly, they show up in the viewport etc. If you use a 3rd party engine, they get nothing.

Exporters see the Redshift materials & their textures, so if you go to Unreal or other packages a lot, it's much more convenient with Redshift.

Basically the advantages of being native material system.

I've written a python Octane->Cinema 4D material convertor, but it's rudimentary & still needs Octane to be installed, so it's a lot less ocnvenient than just using 1 material system for everything.

 

Edited December 6, 2023 by Decade (see edit history)

[Mash]

" Without the denoiser it would be pretty hard to impossible to do almost anything in Octane. "

 

95% of our renders don't use the denoiser. These range from 1k web video elements to 4k 2 minute long product launch videos, from 2k stills for web pages to 16k print res images for packaging. On top of not often using the denoiser, we often need to add extra noise on top of the video in post production. Just because and old geforce 1080 card can't render extreme torture scenes to your satisfaction doesn't mean other people aren't busy getting on with work.

 

"Octane vs Redshift in 2024, which really boils down to a discussion of biased vs unbiased rendering" 

 

I don't think those words mean what you think they mean. There are no more unbiased render engines, they're gone, they dont exist anymore. All current render engines use various tricks to render more detail in certain areas where needed and less detail in others to speed up render times and increase quality.

 

As soon as an engine guides samples towards light instead of uniformly covering the screen, it is biased. As soon as it guides samples to areas of high noise, it is biased. As soon as it blurs caustic rays or secondary light bounces or blurred reflections, it is biased. All render engines are biased, because anything else is a waste of time.

[HappyPolygon]

https://www.creativebloq.com/reviews/octane-render-2023-1

[Jeff H1]

7 hours ago, HappyPolygon said:

https://www.creativebloq.com/reviews/octane-render-2023-1

 

I cringe at CB articles some times.   One of the Cons:  is not easy to install for novices when later in the article they state "installing was typical for a C4D plugin".   They also mention Redshift GPU comes with C4D.  It does not. "It is worth noting that Cinema 4D does come with its own GPU renderer in the excellent Redshift, so I was excited to see how Octane compares from a usability standpoint."   There are some more that are head scratchers.

Outside of that, it's a shame that not enough care is given for the other Octane plugins.  Maya has always been a mess.   Octane for Blender is seemingly in perpetual beta.

 

 

[HappyPolygon]

6 minutes ago, Jeff H1 said:

I cringe at CB articles some times.

 

I wish I had more reliable sources but CB and CGChannel are the only ones I know. And from magazines only one the 3DWorld. 

 

From what I've realized is that all of them are very biased. In other words I strongly believe that they intentionally write non-subjective articles depending what company pays them more to write their stuff. It's the nature of these sources to essentially advertise a software. No one cares writing a good subjective article because no one can. There is no single artist having mastered 2 or 3 software at the same level. In the end if OTOY wants to promote their new renderer it will do so easily by contacting those media handing them a ready-made article with a small fee or a special 3-month license for the author to do whatever he wants. The author will be more than happy to write anything OTOY wants if what he gets makes him feel better (money, personal training, use the offer for his own commercial jobs whatever...) 

 

I think this is also the reason why Blender is much more promoted. Not because the Blender Foundation pays people to promote them but because authors and YouTubers have easier time to talk to more audience (students, hobbyists, newcomers) while being themselves low income or not having high ties to the big players (I mean why write an article if you have a full time job right ? this field isn't exactly in par with media reporters like sport, financial or political. Writing anything apart from these is a secondary income).

 

I cringe at how much 3DWorld promotes Blender and Autodesk. Probably because people reading that magazine are also mostly inclined to those software...