PhotonGI cache

[lacilaci]

Under what sort of circumstances should I turn OFF the 'USE PHOTON GI CACHE', in the Material Output node?

Thanks

Hopefully you wouldn't need to and this option should not be exposed in my opinion...

[Racleborg]

Hopefully you wouldn't need to and this option should not be exposed in my opinion...

Ok, thank you.

[provisory]

Can you post a test scene causing the crash ?

Currently I'm AFK, maybe later.

[Dade]

Hopefully you wouldn't need to and this option should not be exposed in my opinion...

Ok, thank you.

It a safe net if something is not working well. It should be never required (but it is better to have it as last resort).

[frank_yifei]

For a test scene, I tried to make the caustic result look smoother so I increase lookup radius. I found a problem.
using opencl + photonGI, setting caustic lookup radius higher will crash blender. The error message is "ERROR: clEnqueueWriteBuffer
Error: Unrecognized Exception". Path cpu works well with caustic, i can set lookup radius to any value, like 100.
Indirect cache can have lookup radius at any value, under opencl or cpu
My computer have 4 core intel cpu, nvidia 1060 6gb.

Your test scene is working fine here on my Linux+GTX980. Making the radius larger, it will make also the rendering slower. May be, it becomes too slow and the driver kick in and kill the GPU kernel execution.

Given a radius causing the crash, can you try to cut in half (even multiple times) the caustic cache size ? Reducing the size will increase the rendering time again, so if my theory is correct, it should stop crashing at some point.

I'm working on this "larger radius => slower rendering" problem.

I did what you said and it worked very well. Thank you for adding this feature, it is very useful.

[lacilaci]

ok, so came home and tried some latest update aaaand. Is it possible to make coating material not use cache but base material use cache?

[Racleborg]

It a safe net if something is not working well. It should be never required (but it is better to have it as last resort).

Thank you. It's nice to have the option!

[Dade]

ok, so came home and tried some latest update aaaand. Is it possible to make coating material not use cache but base material use cache?

Nope (but why ?)

[lacilaci]

ok, so came home and tried some latest update aaaand. Is it possible to make coating material not use cache but base material use cache?

Nope (but why ?)

coating is usually very low in roughness so material with coating will pretty much always fall bellow 0.2 threshold for using cache. I also seem to have problem with darker textures and cache. Everything is near black in cache... Will post some testscene tomorrow when I'll be able to replicate the issues with small scene.

[Dade]

I have introduced a new stage in PhotonGI caustic cache pre-processing. It merges near photons coming from the same directions. This allows to greatly reduce the number of clustered/overlapped entries in the cache and to greatly improve the rendering speed (especially for larger radius).

This is a rendering without merge phase (0.93M samples/sec):

and this with with merge phase (5.15M samples/sec):

So it is 5 times faster. It is a bit a trade off between caustic sharpness/pointiness and blurring but I prefer the second and the 5x speed increase is worth it.

For a comparison, this is BiDir+Metropolis:

The caustic is sharper but look how much better is the PhotonGI rendered cube with SDS paths (something plain BiDir can not render).