Corona Renderer 1.0 Sneak Peek: Interactive Rendering

[MauricioPC]

https://corona-renderer.com/blog/versio ... rendering/

For those of you that don't know, Corona is a new renderer built for 3ds Max, that now has versions for Cinema 4D (in Alpha) and Maya (still on dev). Currently is about to launch it's first and paid version for 3ds Max (all Alphas versions were free).

You might ask what is Corona ...

Corona is ... both biased or unbiased renderer. It does not follow the traditional division and tries to combine the best of the two worlds - the speed and flexibility of biased renderers with the image quality and simplicity of unbiased ones. As such, it supports both unbiased and biased rendering modes, that can be additionally combined together to get the best bias/noise ratio.

Corona is ... entirely CPU based with no intention of making a GPU version in the foreseeable future. The reason is that the CPU version is fast enough, and there would not be nearly enough speed improvement to justify the amount of changes required and reduced flexibility.

Corona has ... this minimal requirements for the two versions of Corona - standard and legacy. The standard one uses the SSE4.1 instruction set to achieve maximum speed on modern CPUs, and needs an Intel i3/5/7 CPU or newer Core 2 (e.g. 9xxx). Legacy version should run on all CPUs of the last decade, but is about 20% slower. The installer automatically chooses the correct version for you.

Also of note that with it's launch, the owners/devs of Corona are trying something they call "Fair SaaS". But as a option, they also have the license perpetual purchase.

There's a nice post below where they addressed concerns about their users/clients who didn't like some things about the price structure and made some changes to it.

https://corona-renderer.com/blog/discus ... litics-ii/

Moderator edit: edited the thread title to actually include the product name - HB

[Hirazi Blue]

While SaaS for a renderer makes a lot more sense than for a fully equipped DCC (in light of the potential "hostage situation"),
the idea of "Fair SaaS" IMHO is a bunch of baloney, Orwellian "newspeak" if you will:
If it smells like SaaS, feels like SaaS and tastes like SaaS, it's probably SaaS...
(with a big nod to the old Cheech & Chong “dogsh*t” bit)

[MauricioPC]

Of course SaaS is better for the developers as a steady income. But they still try to be more transparent and open to suggestion. And they still have a box version with/without subs.

SaaS

Unlimited access to all major, monthly, and daily experimental builds.
Costs 24.99€/month (19.99€ for first year with our initial promo discount)
1 floating workstation license (usable also as render node) + 3 floating render nodes
Paid monthly, with voluntary option to pre-pay 1 year
Cancel at any time without any charge
NEW: we added SaaS versions with 5 render nodes instead of 3 (29.99€/month), and 10 render nodes instead of 3 (45.99€/month)

Box

Forever access the version 1.0
Costs 449€/ paid once (349€ with our initial promo discount)
1 fixed seat workstation license (usable also as render node) + 3 fixed render nodes
Access to all bug fixes for major version 1.0.
Eligible for future paid upgrades to versions 2.0, 3.0, …
We have scrapped the 90 days of free subscription – no longer applicable
Subscription
Costs 99€/year
Gives permanent access to future major versions (2.0, 3.0, …), monthly builds and daily experimental builds released while the subscription is active
Must be purchased when buying the box license.
Valid for 12 months, can be renewed

As an example ... Cebas launched the new Thinking Particles v6 plugin for Max. It looks amazing, but they scrapped the perpetual license option and now you can only rent from 1 year periods. No options and as of now, no hearing of users.

Having at least one company hearing users about it is better than nothing. And Corona is coming for C4D, for those who want AD to just crash and burn!

[Hirazi Blue]

Don't get me wrong: I can clearly see they are trying something, potentially interesting in this day and age.
my remark was solely directed at the semantics of naming it "Fair SaaS".

[MauricioPC]

Don't get me wrong: I can clearly see they are trying something, potentially interesting in this day and age.
my remark was solely directed at the semantics of naming it "Fair SaaS".

Yeah, I agree.

And talking about the renderer itself, it does look very fast. I know some guys here using it on commercials (since the Alpha was free for production as well) and saying it is very reliable.

[FXDude]

Wonder how much faster than traditional path CPU tracers, it does look very fast though. (or also vs Redshift on say a GTX780?)

But keep in mind that sometimes getting rid of the noise beyond 80%-ish of visually acceptable levels can take a long time.

And the inherent denoising of video compression when playing screen recordings is often a bit deceiving.

Making things look almost done while they are only almost done for previews.

But it does look pretty quick

[Nox]

Basicly Corona is unbiased path tracer (like any other) but based on Embree (~20% speed boost) and uses something similar to Light Cache to calculate a lot of secondary bounces.

It's slower than biased render engines like Vray (except some specific cases) or Redshift but faster than over 9000 unbiased clones popping out every week.

Great thing about Corona is almost zero setup time, you just put lights, create mats, press render and wait, thats all.

But currently in area of handling huge data sets it's nowhere near Arnold, Vray or even Redshift (if you need that).

[FXDude]

Also here is a page with some benchmarks..
https://corona-renderer.com/forum/index.php?topic=559.0

Using a sample scene, the range was from 4m to 12min for more than decent results, which is indeed absolutely excellent!

Yet it also seems that the sample scene is somewhat optimized to the point where people thought it was somehow faked (but it's not).
and you can see that it's borderline okay in the (noisier) shadows.

And also, searching youtube, I found but a couple of animated sequences, one of which was quite a bit flickery (with light caching), and the other was pure pathtraced (with no rendertime revealed).

To date, I only mostly saw very nice Arch Interior scene stills, (and a one very nice grassy exterior sill)
but it may be a question of setting things right for animation.

And to add to the list of sometimes deceiving things about perceived performance,
screenrecordings are also often downsized, or at least viewed downsized, so it's often important to see final rendertimes and final images, such as in that benchmark page with images.

But Thanks for that great insight Nox!

[Nox]

Well, those 4 min. images have lots of noise everywhere. Final render of this scene is on main page (68 minutes).

[rawalanche]

Well, those 4 min. images have lots of noise everywhere. Final render of this scene is on main page (68 minutes).

Just to set some things straight. Rendertime for final image was indeed 68 minutes. BUT:

1, The final was rendered by me on very outdated CPU (i7 870) which is about 50% performance of modern haswell machines (i7 4770k), so that puts us to around 34 minutes.

2, Final render was rendered at 1920*1080 res while benchmark scene runs only at 1024*576, which is almost twice as small resolution, meaning almost 4 times less pixels = 4* faster render.

3, Final render ran for 150 passes, while benchmark scene runs only for 100 passes. Sure it's not perfectly clean. It's just to demonstrate you can get reasonable quality image at reasonable time.

But best thing of all is that you don't need to take my speculations for granted! You can super simply download benchmark scene (it does not need 3ds Max installed), edit the conf file, and set pass limit to 150 passes and resolution to 1920x1080. Then you can very simply try for yoursef, how it really runs on your machine.