Struggling with transition

[AndreasResch]

One more question - maybe it's a bug. Why does my rendering only go about half the way when using "Out of core" rendering?

Here are the settings.

[Dade]

Notice the "Samples per pass: 512" label: you are probably rendering less than a single pass (halt condition is less than 512, use an halt condition that is a multiple of 512).

[AndreasResch]

I fear I don't understand. The halt condition IS 512 samples.Where did I set it to a value less than 512.

[Dade]

I fear I don't understand. The halt condition IS 512 samples.Where did I set it to a value less than 512.

Are you talking of PATHOCL or TILEPATHOCL ?

[AndreasResch]

There was one question about tiled rendering being more noisy than non-tiled rendering and there was one about "Out of core" renderings not being finished. They are unrelated I'm not sure to which of those questions you are referring to with your answer. But as you talk about "512 samples" I assume you answered the "Out of core" question. And as you can see in the screenshot, tiled rendering isn't enabled there. So it might be PATHOCL - although I have CUDA enabled, so I'm nor sure if that applies either.

[Dade]

There was one question about tiled rendering being more noisy than non-tiled rendering and there was one about "Out of core" renderings not being finished. They are unrelated I'm not sure to which of those questions you are referring to with your answer. But as you talk about "512 samples" I assume you answered the "Out of core" question. And as you can see in the screenshot, tiled rendering isn't enabled there. So it might be PATHOCL - although I have CUDA enabled, so I'm nor sure if that applies either.

In that case, I assume, you are doing CPU+GPU and that is the source of your problem: CPU and GPU works independently, you need to use only the CPU or only the GPU to be able to just one single pass. If you use both you will stop the rendering at some point half way (the halt condition check the sum of CPU and GPU samples count if you use both).

[AndreasResch]

That's all pretty confusing and it's no wonder that it takes so long to get into good results. So the bottom line is to stay away from "Out of core" as you will never know how much samples you will get in the end. And eventually one half of the image looks different from the other.

After looking at the same scene in Cycles it seems as if Cycles automatically accesses the shared memory once the GPU runs out of memory. So this scene (together with the rest of what's running on the OS at that time) ends up using over 7GB of GPU memory although I only have 6GB of dedicated GPU memory. But it works so flawlessly, that I didn't really realize it.

It seems as if I will have to wait for V2.6 until a few of those speed bumps are ironed out or I decide to get myself some stronger hardware. As it is now, it's too time-consuming to come up with individual solutions for each scene.

Cheers for the help though.