Hi all, was looking into VR performance tips for foliage and I learned about the DitherTemporalAA node which Epic recommends for translucency. I ran some tests before integrating into my project and I am seeing that using the node actually hurts me more in terms of performance. I attached my tests with and without the node in my material. Any ideas of what could be happeneing? The scene I am working on is in a garden with plenty of trees and bushes so any other tips would be appreciated.
Every graphical feature you use has a cost, every node you add to your material makes it render more slowly. In this case it is sampling an extra texture for the dither pattern and doing some additional math. Plus, more pixels will be rendered to twice because any time the dithering causes a pixel to be clipped, the grass behind it might need to be drawn. You will probably see more red in the shader complexity view to reflect this.
I see what you mean, but doesnt that go against what is said in this image? Isnt the point of dithering to render less and make it look like its transparent to reduce overdraw? I understand this node to cheapen the transparency effect, not increase the performance required to render properly.
The point is to make it look better and have fewer artifacts, not run faster. It is potentially cheaper than using alpha blending, depending on your geometry and materials (triangle order is particularly important, and whether they can be sorted front to back), but it is not cheaper than using masked rendering by itself, and certainly not cheaper than just rendering opaque like in the left side of that image. When you hear about something being faster, you need to ask “compared to what?”
The reason you are not seeing any performance gains is due to the amount of translucent objects you are trying to render. As OwenWP said this node is to make transparent objects look better and not run faster.