Max Samples Question

[treddie]

In the features list for 2.58e, it was stated, that RS had "Fixed rendering more than 6400 samples via command line."

I was wondering if this was a typo; Should "6400" have been "64000" as the slider shows?

At any rate, when I try to use the command line to do a -s 500000, it simply drops me back to 64000. I also remember a post saying that since you cannot have an infinite maxsamples (because a GPU does not have infinite memory), is there some maximum ABOVE 64000? I ask this since if 64000 is the maximum limit, then even with a low poly scene, some backlit objects simply cannot clear up within the time it takes to complete the render. The only other option is to run the render roughly 5 times, then combine in Photoshop. It would be nice to be able to run beyond the 64000 limit if this is feasible with the GPU cards out there.

[abstrax]

In the features list for 2.58e, it was stated, that RS had "Fixed rendering more than 6400 samples via command line."

I was wondering if this was a typo; Should "6400" have been "64000" as the slider shows?

At any rate, when I try to use the command line to do a -s 500000, it simply drops me back to 64000. I also remember a post saying that since you cannot have an infinite maxsamples (because a GPU does not have infinite memory), is there some maximum ABOVE 64000? I ask this since if 64000 is the maximum limit, then even with a low poly scene, some backlit objects simply cannot clear up within the time it takes to complete the render. The only other option is to run the render roughly 5 times, then combine in Photoshop. It would be nice to be able to run beyond the 64000 limit if this is feasible with the GPU cards out there.

Hi Treddie,

the 64000 samples/pixel are pretty much the limit. It's a hard limit. To get around it, we would need to split the total samples into 64k blocks and merge them, which isn't implemented yet. You can already do that manually yourself though.

On the other hand 64k samples/pixels should be enough in most cases, especially using PMC. Could you post an example where that isn't the case?

Cheers,
Marcus

[matej]

Marcus is back??!!

YES!!

[treddie]

Marcus, thanks for the reply.

I am in the process of getting that together to post. I hope to have it up today.

[treddie]

Here are images of both the render, and the setup (nodes, nodes settings and an image of a pushbutton wireframe to get an idea of the poly counts. To try to test for speed vs. poly count, this version has a relatively low number of polys yet still takes 22 hrs to finish rendering at maxsamples = 64000, but only because of the pushbuttons. Everything else in the render clears up, noise wise, very quickly. At the end of the render, the pushbuttons still have a long way to go. The black panel surface might look like it is noisy too, at first glance, but it is actually a ridged fractal texture:

I am running a batch job of six renders to composite in PShop to get the noise reduced. I know there is a procedure for dealing with noise that was posted some time ago, but this test was already setup and in the process of being rendered. I have the nagging suspicion that that procedure would not work in this case if only because the noise removal rate on the buttons is so poor.

Here is the setup:

One item of note...I used a material mix to get the unlit button color to be fairly intense. This was only done when I was still working with a subsurface scattering approach, because subsurface would not allow me to get a really bright colored plastic texture. Since I have since nixed the subsurface idea completely, I MIGHT be able to dump that extra "layer".