Here's a surprise. I tested this bucket with a combiner material texture and also a "baked" decal version.

Click to view attachment

The Baked version rendered in less than half the time of the Material version in a regular render (as I expected), but the advantage disappeared when I used Multipass.

Some numbers:

Click to view attachment


Anyone know why this is?

That should just be the overhead of all the other things that are included in a render. I think you'll see a greater difference when the materials are more complex and reflections are used.

Yup. As an axiom to this corollary (...umm...probably vice versa) - Try a test with hair - and watch the difference SOAR. No multipass wins by far. It's very very fast. (I'm doing 4-6 secs a frame for my 11 sec entry for final No MP, 15e)

But remember I'm just talking about baked textures. There is no way to bake hair into a decal as you can most surface parameters*.

I tried a more complex material with combiners inside a combiner...

Click to view attachment


and baking does start to pull ahead, but just a little.

Click to view attachment


So i'm still wondering why baking is so much less of an advantage when using multipass?




*It turns out that Baking doesn't include a reflectivity map so this test is with Reflection turned OFF.

Great 'chartage' Rob!

perhaps the material is not recomputed for each pass ? ie makes use of some data from pass 1 computation?

(I knew you were just looking at baking...was just pointing out another case of speed up using NO MP)

possibly, but as far as I can tell the passes all take the same amount of time.

And 4 passes has about the same time disadvantage as 64 passes does so that doesn't suggest that something is being done once for the benefit of future passes.

ok then possibly the anti-aliasing pass for No MP recomputes the material ?

Or obviously in some way the pipeline of computation is less efficient and different for No Multipass?

That might be why decaled bump maps don't show on a surface that also has a material applied to it when using No MP, but do show up when using MP.

Here's another way to look at it

Click to view attachment


the biggest difference between Regular render and Multi-pass is that no anti-alias operation happens with multi-pass (the anti-aliasing is had by averaging all the passes together)


So, my guess is that whatever A:M must do to anti-alias a material in a Regular render is quite a bit more time consuming than what it must do to anti-alias a bitmap representation of the same texture in a Regular Render.

But Regular Render is still faster overall for Materials unless you cut your multipass down to 4 or fewer passes, which would get you not much anti-aliasing.

Where do you read that the material renders faster? It is less fast going after your chart no matter what, even so the advantages does cut down quite a bit, or are we talking about higher decalresolutions?

*Fuchur*

Regular Render is faster for Materials than Multipass is for Materials. That's what I mean.

Materials are much slower in Regular render than decals, but they are still faster in regular render than they are in multi-pass, unless you just do 1, 2 or 3 passes.

My understanding is that Multipass records all the parameters... EVERYTHING*... EVEN THOSE THAT WILL NOT BE USED... up to the number of passes specified. When it hits that upper threshold it quits sampling, storing and calculating and renders with the data that it can use.

This is why adding and subtracting various options while using Multipass yields similar rendering times.
This is also why adding options while using Multipass will get those rendered out in the same amount of time for 'free'.

(This has been an unlearned and naive description of the process and betrays my low level of understanding)



*Everything designed to be calculated into Multipass. Some things aren't included in the Multipass Set.

Not to change subjects, but I've been noticing my displacement maps don't show in the regular renderer...but they DO in multipass. Anyone else getting this? (V15.0G PC)

Haven't tried it with 15h yet, but bump maps disappear in a regular render when the same surface also has a gradient material applied. But with multipass, they show up just fine. Haven't done much with displacement maps recently.