Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Messages - jantru

Pages: [1] 2
I've been trying to figure out if there is a way to improve the performance of the Tile Sampler and found out (correct me if I'm wrong) that the performance of the Tile Sampler is mostly (negatively) affected by multiple Sampler nodes it uses.

I also noticed that the Sampler node computation time is directly connected to the resolution of the input image. I did some testing and found out that with a 256x256 resolution it takes something over 1ms to sample grayscale image. But with a 4K image, it already takes almost 20ms. So, having multiple Sampler nodes within a graph (as the Tile Sampler has) can probably result in quite a performance hit.

However, there is a catch. Even though I gave the Tile Sampler low-resolution inputs, the computation time was still pretty much the same as with high resolution. So, I looked inside and found out that internally the inputs get upscaled to "Relative to Parent" resolution, so in the end, it computes with a high resolution anyway. And that I think is an unnecessary performance loss.

So, by slightly modifying the Tile Sampler and adding a few parameters I was able to control the resolution and fix the problem (kind of).
Here are some images of what I did:

It's obviously not the ideal solution to the problem but I think it can at least serve as a "proof of concept" that there are some possible ways of improving the performance.

The thing is that usually, it's not necessary to use high-resolution input images for masks, etc. When I use the Tile Sampler to scatter some rocks for example It's enough to use 1K, 512, or even 256 resolution for the mask. When working with high-resolution materials it usually doesn't matter if a rock is exactly on THAT pixel or is a few pixels to the right for example.

So, the bottom line is that using lower-res inputs can potentially improve performance while giving practically the same results. Of course, if there is only one Tile Sampler in the graph the difference won't be noticeable, but if there are like five or more you can feel it's more responsive. Especially while working with high resolutions.

I don't know how accurate the timing labels inside the Designer are, but anyway, here is a comparison between the original Tile Sampler and my modified version (or rather optimization ON/OFF):

What do you guys think? Do you have some other performance tips? :)

Eventually, I would like to thank Substance for this amazing community contest. I had a blast! Even though it wasn't always easy :)

Also thanks for all the feedback and support. Good luck to you all! :)   

Hi guys!
I haven't slept for 48 hours, but it's finally finished!  ;D
Here are my final renders:

Another render, this time from Substance

After all the hours of painting, I'm finally rendering...

painting process is almost done. :)

Good job!  I like the first one better, but it's just the way I see it :)

Looks good. Keep pushing, I would like to see this one finished!

Thanks! :)
Im pushing, it's almost two in the morning now. :D

Oh wow, loving the style!! Cant wait to see it finished :) Keep up the awesome work

Thanks a lot! I really appriciate it! :)

Pages: [1] 2