Unexplainable Render Time Spike Tweet

[prisonbread]

Hey, folks. I have a scene that has no animation, but camera animation and some pulsing lights (blackbody textures). It's a pretty simple scene, and all of these pulsating lights, though not moving, have octane object tags on them just to be sure. I have also, for saving time during the export phase of each frame, have it set to manual object update. Anyhow, things were moving along for a while at about 10 minutes a frame, which is reasonable considering I just have one GTX 980 ti. But eventually I noticed the render times jumping between three general lengths of time though the information the screen is roughly the same. They go from 9-10 minutes, to one frame that is 21 minutes to render, back down to 9-10 minutes, and finally have plateaud at a whopping 37 minutes a frame. I've got some included images and you can see on the right the corresponding render time for each of these images. I am really curious what could cause this arbitrary jump to this render time, is my GPU overheating after frame 25?

[bartolos]

Actually I had a similar thing. I had a scene where I had some lights turn on at some point. I used animated Power parameter for this. The frames where the lights were dark rendered significantly longer than when they were at their full power. It was before adaptive sampling was available so it didnt have anything to do with noise levels.

It's not a plugin problem though, but the render engine. Probably something involving importance sampling between light sources gets messed up when there are dim lights.

B.

[prisonbread]

Wow, Bartolos, I think you're on to something. I checked out the sampling rate of this blackbody material that makes up all of the lights in the scene, and I must have, when experimenting with noise reduction cranked it up there when there was just a single on of the lights. So, basically I have like 16 lights all with a sampling rate of 1000, also reaching a power of 200 at the peak of the pulse. I will reduce these rates and see what I get. Thanks for the suggestion!

-Ben

[bartolos]

I think the sampling rate is not a direct parameter (like a literal number of samples taken for a light) - but rather a kind of importance priority. That's how I understand it, because Octane is theoretically unbiased...
What I mean is when there is the same value on all the lights it means they get equal attention when sampling. So it doesnt matter if it's 0.1, 1, 100, or 1000, as long as it's the same value across all the lights in the scene, they get sampled the same, so it should have absolutely no impact on render time, but rather at how quickly the noise gets evened out...

Say - you have a room scene with soft lighting, and a bright tv screen. Most probably the soft light coming from big area lights will not be noisy, because the samples get an even distribution. But the light from the TV will need special attention. With no sampling priority control it would end up noisy, because it is a "spiky" lights source compared to other area lights. This is where sampling rate comes in.

Am I right?

Anyway, as I said it shouldnt affect render time. So there's something else going on which makes the rendering slower.

From old days I did signal processing there was a thing with FPUs being MUCH slower when calculating denormal numbers (very small values, like 0.0001) which maybe is the same with GPU. Like having very dim lighting causing it to slow down.

B.