Path Tracing always as Secondary GI Solver

[Mr.White]

I've been wondering over the last few days if it makes any sense to always use Path Tracer as the secondary GI solver. I am aware that in some scenes it is better suited (few bounces / outdoor scenarios) and in others not (more bounces / interiors)

The idea was that I always want to have as unbiased a solution as possible. I accept longer render times for this.

Now my question: Is it “worth it” or are the differences so marginal that it makes no sense? Are there any experiences or opinions here? I would very much appreciate feedback here. Thank you!

[Abdullah_Alallah]

TBH i would love yo hear more of the process of that since i use 4K and path tracer, like how much of a difference in accuracy would you gain in using 2 Path tracers. i understand that it's scene-dependent, but are you gaining accuracy at the expense of more time in the render only?

[maru]

My advice would be to stick to defaults. You can get a really tiny bit of extra realism, but you risk adding lots of extra time to your rendering.
In Corona, using the cached GI doesn't come with some visible drawbacks like back in the old renderer times, you don't end up with things like missing contact shadows.

[pokoy]

If time is not a concern I'd stick to PT/PT. Even on exteriors, caching might result in a less detailed GI. If you don't know the difference you might not notice but from all side-by-side tests I've done PT/PT just looked better.

On interiors, I had a few cases where both secondary cache solvers would result in splotchy areas no mater what I did. While the 4K cache always produced better results, last time I looked there was still no documentation for it available so if you need to to tweak something you really don't know what the parameters do.

[maru]

I agree with PT+PT for exteriors.

[Mr.White]

Thank you for the input - really appreciate it! Good to hear some opinions on this matter.

[romullus]

Speaking about bias, how about MSI? I was quite shocked another day to find how much light it could prevent to reach the areas that doesn't even seem that much occluded. Attached are comparison of default MSI 20 vs MSI 0 (off). Of course i would not recommend to turn MSI off in your scenes, because you'll surely get a ton of fireflies and much longer render times, but it might be useful to try adjust it in some darker interiors and see if you like what you see. Does anyone remember the old MSI 200 rule? :]

By the way, new firefly filter does pretty amazing job at removing excessive fireflies caused by turned off MSI (see the last attached example).

[dj_buckley]

Speaking about bias, how about MSI? I was quite shocked another day to find how much light it could prevent to reach the areas that doesn't even seem that much occluded. Attached are comparison of default MSI 20 vs MSI 0 (off). Of course i would not recommend to turn MSI off in your scenes, because you'll surely get a ton of fireflies and much longer render times, but it might be useful to try adjust it in some darker interiors and see if you like what you see. Does anyone remember the old MSI 200 rule? :]

By the way, new firefly filter does pretty amazing job at removing excessive fireflies caused by turned off MSI (see the last attached example).

Really interesting test, MSI - 0 really brings out the translucency of the leaves, yet it seems the firefly filter knocks that back again despite there not really being many fireflies on the leaves

[arqrenderz]

WOOW the change is indeed huge, do you have the render times for those ?

[Aram Avetisyan]

As it is said, Corona is (un)biased renderer, MSI being one of the main biasing factors. You can see crazy-beautiful things happening with caustics when there is no MSI and there are strong reflective surfaces (keep in mind long render times). I remember forgetting to set render limit for a volumetric caustics scene, and with >1000 passes rendered, beautiful caustics appeared, in places I would not think of them appearing.
I agree that with firefly removal MSI can be tweaked with more confidence. Yet again, be sure you know what you are doing when tweaking the defaults.

[maru]

https://support.chaos.com/hc/en-us/articles/4528471000849-What-is-highlight-clamping-3ds-Max

[romullus]

WOOW the change is indeed huge, do you have the render times for those ?

About 3 minutes for each, it was set to pass limit as stopping condition. I didn't look at noise percentage, but MSI 0 was obviously much noisier before denoiser. One would need to make more in-depth tests before drawing any conclusions, but as Aram said, with the new fireflies filter you can now play with MSI. Just a reminder - you want to increase it to make its effect less destructive. Setting it to 0 is a special case, i.e. MSI is effectively turned off, which is not recommended.

[romullus]

https://support.chaos.com/hc/en-us/articles/4528471000849-What-is-highlight-clamping-3ds-Max

I never liked highlight clamping, but maybe i need to give it another chance someday.

[davemahi]

I am a big fan of the PT + PT approach as well. It's 2025, I don't want to be caching GI for offline rendering any more haha. I do change a few settings like highlight clamping, but pretty much never change from PT.