Weird Streaks

[treddie]

Has anyone noticed this before?

This scene is my version of the proverbial light diffraction experiment, still not dialed in yet. Along the way I noticed this problem. I thought it might be related to scene scaling, but even if I scale it up by 10, same problem. The model was built in inches, imported into C4D in inches, and exported in inches at document scale =1. Octane import is set to inches as well. The prism is 2" x 2" on each side.

The streaks are just plain weird. There is no background sky or HDR environment...just a little cube emitter off to the top left of the scene, out of the frame.

[roeland]

The noise also shouldn't be there. What kernel did you use to render this image? Do path tracing and PMC yield the same result?

--
Roeland

[treddie]

Well, the noise I think is because I didn't let the render complete, because of the streaks. So I aborted the render. This is with path tracing. I tried PMC but for some reason for this project, I cannot lower the power of the background; it stays overexposed at the same brightness. At any rate, it does not cause the same streaking.

[treddie]

I'm sorry Roeland...The reason PMC was getting so over-exposed was because I somehow had moved my exposure and ISO sliders hard right. PMC does NOT have the same streaks problem as PT. Octane says that it needs to render for 9 hrs which seems odd since all the parts are very low res:
Prism, 8 triangles
Card, 12 tris
Floor, 12 tris
Laser tube, 204 tris
Emitter surface inside tube, 28 tris

With PMC, the prism is taking forever to develop. In another thread, Marcus pointed out to me that if an emitter is too small, the chance of paths getting to it is severely reduced. In this render, the light emitter cube lighting the scene is only .25 inch on a side. Perhaps that explains why the prism glass is practically black right now, and everything else is almost finished after only 23 minutes. Tons of fireflies, maybe for the same reason?

[treddie]

Geez. Still a newbie in the middle of the learning curve. All I had to do was kick up my sampling rate on the little emitter cube.

This is at 3 min of a 10 hr render:

So PMC does not have the streaks issue, and PT does for some reason.

[roeland]

The laser emitter should be set to more power than the cube for its beam to be visible. Is it clearly visible if it hits the plane directly?

The light passing through the prism is not rendered with direct light rays, so changing the emission sample rates won't affect how fast it is rendered. The sampling rate of both emitters should be set to 1.0 .

If you change to path tracing without changing any other settings, what does the rendered image look like?

--
Roeland

[treddie]

That was it! Early on when I was setting up the laser intensity, I couldn't get it to work. So I kept upping the power and efficiency and when THAT didn't work, I maxed out sampling rate, too. It was the whacked out sampling rate that was causing the streaks. I know not to do THAT now!

So here is the latest render at 21 minutes. I dropped the scene lighting down so the laser light would be more visible. It seems odd though that even though I dropped the laser's s-rate back to 1.0, that with power and efficiency still maxed out, it should be burning a hole through the card by now. You can juuuuust start to see a few pixels of the laser light starting to hit the card.

[treddie]

I built two images and each appears to still have issues. The one on the left is rendered with PMC and the one on the left, PT. In order for the PMC version to work, I had to drop the wattage of the scene light from 100W (for PT) down to .1W (for PMC). That seems wrong. I have other scenes that can be rendered back and forth with no changes. I have gone back over my settings and as far as I can tell, everything is OK. My only other option is to reboot the system, if the problem lies in the GPU. Unfortunately, I cannot reboot yet as I am testing a computer program that needs to run for about 5 days.

The callouts talking about odd cutoff of the light dispersion I think may be incorrect...I think it is because the spot on the card as reflected inside the prism is being cutoff by one of the prism edges...The other side of that edge is reflecting the floor:

[roeland]

Scenes like this are hard to render for a path tracer without bidirectional path tracing, like Octane. You will get the best result with PMC. There shouldn't be such difference in colors between PT and PMC though, If you send us the scene we can check for any problems with Octane.

The odd cutoff may be a reflection of the card in the ground plane of the prism.

--
Roeland

[treddie]

Yes, I think the cutoff area is a reflection of the card, too. I think the prism vertical edge closest to the card and the faces to either side have been reflected inside the prism multiple times and ended up on the bottom face of the prism. One side of that edge is reflecting the floor and the other is reflecting the card.

Also, noticed that the laser is not reflecting in the front face of the prism even though its reflectivity is set to .5.

One other issue I have noticed is that if I save out an ocs file and do not have a background node (Daylight or HDR) connected to the render target, then when I reopen the file later, Octane acts like it is trying to render a background anyway, even though it is not there. I have to connect up a background node and then disconnect it, before I start rendering to get it to work right.

Here is the scene and mesh in a zip. I left out the HDR image because the site will not let me attach a zip large enough to carry it. But it's not important anyway.

Tom's Prism.zip

(7.55 MiB) Downloaded 154 times