Displacement = Greatly reduced Samples per second

[3dreamstudios]

I have a scene that I am able to get around 5-10 MS a second depending on if I have both my cards engaged. If I bring in my mountains created with a height map from http://terrain.party (yep just found this, looks like a great recourse I'll use more) The samples come down to nearly .15 MS a second. That's a drastic difference.

If I have the mountains all by them selves in a scene I get 8-16 MS which I would expect. I've attached two images, one of the mountains the other of the scene I'm trying to put them in. Any ideas as to why this difference....I would expect less with them together but not 30 times less.

[geo_n]

That's a very known issue with displacement. I posted a lot about this in the past. I thought it was my maxwell card.
But with newer builds optimizing maxwell card the displacement issue didn't get better so I think its an issue of the code itself.
Its better to use instances for grass rather than displacement.

[3dreamstudios]

That's a very known issue with displacement. I posted a lot about this in the past. I thought it was my maxwell card.
But with newer builds optimizing maxwell card the displacement issue didn't get better so I think its an issue of the code itself.
Its better to use instances for grass rather than displacement.

Hmm...I bet there is something in there causing a bottleneck of some sort. As both scenes by themselves render quite quickly on my machine. But when put together (add displacement to the main scene) it CRAWLs!!!!

As for instances and grass....I'm just wanting the mountains...forget about foliage, grass, rocks etc..etc.. Although Octane would handle that with no problem from what I've seen so far. Have had 10 of thousands of trees in scenes and it rips through it like butter. I mainly use DL on my exteriors but with diffusion sometimes... Anyway.

I wonder if this could be confirmed, or maybe already was somewhere in the past? I'll wait for Juanjgon to comment. Hopefully he will see and respond. Thanks!

[juanjgon]

It is hard to say. I was thinking about it, but really I don't know why you have this drop in the rendering speed. Perhaps the problem is that mountains model is too big in relation with the house model, and the Octane ray tracing acceleration structures are not working fine.

Can you please try to use a model for the displaced mountains with the same size than the house model to see what happens?

-Juanjo

[3dreamstudios]

It is hard to say. I was thinking about it, but really I don't know why you have this drop in the rendering speed. Perhaps the problem is that mountains model is too big in relation with the house model, and the Octane ray tracing acceleration structures are not working fine.

Can you please try to use a model for the displaced mountains with the same size than the house model to see what happens?

-Juanjo

Do you mean same physical size? Of course the mountains I put in the scene are quite a bit larger than the size of the building.

When I get some time I will do some more testing. and post my findings to maybe help you look in the correct spot of the code...

[juanjgon]

Do you mean same physical size? Of course the mountains I put in the scene are quite a bit larger than the size of the building.

When I get some time I will do some more testing. and post my findings to maybe help you look in the correct spot of the code...

Yes, the physical size. Try a scene with the mountains object only a little bigger than the house object.

-Juanjo

[3dreamstudios]

Do you mean same physical size? Of course the mountains I put in the scene are quite a bit larger than the size of the building.

When I get some time I will do some more testing. and post my findings to maybe help you look in the correct spot of the code...

Yes, the physical size. Try a scene with the mountains object only a little bigger than the house object.

-Juanjo

Didn't see this post until I did a few tests this morning. What I found out is interesting.

1. Octane Displacement is a bit different than LW's as far as how it reads/uses the height map. This is fine I guess..just interesting.
2. When Octane is doing the displacement it seems to reduce the render time by quite a bit. I guess this makes sense to some degree as there are quite a bit of calculations being made on making the actual displacement...it's not just geometry. Fine.
3. When LW does the displacement Octane actually can use it because it's able to load the geometry into the GPU and then use as regular geometry. It does not affect the render times (in these test scenes) much at all. It's just more geometry...and we know Octane chews threw stuff normally.
4. Octane Displacement must be needed to get it's SMOOTHING from OGL or something as smoothing is not working properly. Even with Shaded Textured On in OGL there is nothing for it to pull from because the Displacement happens at render time...so all you see is a flat plane in OGL.
5. Smoothing with LW displacment works great as expected when rendered inside Octane.

I will try another scene with displacement similar size to the rest of the scene but this is not a good solution for this situation of course. Honesly would just use LW displacement so I can see the mountains, fake their size etc..etc...

PS. I tried the subdivide option in Octane Displacement on all levels to see if Smoothing was any better. It was similar results on each.

[3dreamstudios]

Do you mean same physical size? Of course the mountains I put in the scene are quite a bit larger than the size of the building.

When I get some time I will do some more testing. and post my findings to maybe help you look in the correct spot of the code...

Yes, the physical size. Try a scene with the mountains object only a little bigger than the house object.

-Juanjo

Ok this test was with exact same scene as before and as you can see, you were right it seems something goes way wrong with the optimization or something when the size of the displacement is much larger than the rest of the calculated scene. That's way over my head as to why...but maybe that can help you...

Same scene, same displacement just scaled to be similar size to my building site. Very little difference in samples per second. Just like in LW test above when LW did the displacement.

[acc24ex]

PS. I tried the subdivide option in Octane Displacement on all levels to see if Smoothing was any better. It was similar results on each.

regarding displacement smoothing , saw somewhere you should use 32 bit images for better performance - try that out.. unless we're talking about poly smoothing which, yeah, you cannot control it after it is displaced.. displacement doesn't take into account number of polys, treats it like on a per pixel basis, your poly area divided into resolution of the bitmap / something

[juanjgon]

Ok this test was with exact same scene as before and as you can see, you were right it seems something goes way wrong with the optimization or something when the size of the displacement is much larger than the rest of the calculated scene. That's way over my head as to why...but maybe that can help you...

So as you can see the issue is that render a very big displacement object in a small scene is not good for the ray tracing performance. The problem can be a poor optimization tree structure for the ray tracer due to the big amount of not used space in the scene and the displacement algorithms ... anyway I don't know the details of the Octane ray tracer architecture.

Perhaps for this kind of projects can be better use the LightWave displacement, loaded as you say as a dense triangular mesh, if you have enough GPU memory.

-Juanjo