Corona 1.01.00 CoronaBitmap slower than Max's Bitmap

[Bormax]

Hi,
Today I installed Corona 1.01.00 and tried it on scene I'm currently working on. Unfortunately I've noticed that Corona 1.00.02 DB 2015-06-16 1.01 RC2 rendered my scene a bit faster. I read that using CoronaBitmap should speed up rendering process and I converted all Bitmaps to CoronaBitmaps using Corona Material Converter. Render time grown up for 26%.
Statistics are in attachments.
Also I've done the same with another scene and I've got 12% longer render time with CoronaBitmap instead of Max's Bitmap.
DUAL Intel(R) Xeon(R) CPU E5-2660 @ 2.20GHz, 32GB of RAM

[Ondra]

hi, can you try the comparison without those trees?

[Bormax]

Sure.
Forest (600 trees), bushes and grass (about 800K tufts of grass (not visible from this point of view)) are made using Multiscatter, plus some trees are corona proxies manualy placed. Stones are scattered by CoronaScatter.
Here are the statistics without trees, bushes and grass. Pass limit raised to 40. When I first time redered after conversion parsing took 10sec, here is statistic about 2nd render.
Some strange thing happened with the bottom of the roof with CoronaBitmap version. You can see this place closer on attachments with Bitmap and CoronaBitmap. BerconTiles+MultiTexture used here to texture this surface.

[Ondra]

is it possible there is duplicated geometry (2 coplanar faces) under the roof?

[hubrobin]

Hey Bormax,

this doesn't look too good indeed, but I would need the data to make some sense out of it. Would it be possible to get your scene? You could upload it privately through 'https://corona-renderer.com/upload/'.

Thanks!

Robin

[Bormax]

Hi guys!
Ondra, I've checked geometry - there are no coplanar faces in that place.
Robin, this scene is quite big and it includes xrefs (buildings on background). Do you need whole scene or I can save some part of this scene, if you have any ideas what objects can cause these problems?

[maru]

I would guess it's bercon/multitexture problem.

[hubrobin]

Hey,

we could just start with that roof problem. Maybe it will give some ideas about the slow down. So if you could just send me the roof.

Thanks,
Robin

[Bormax]

Yes, Maru is right - texture problem caused by combination of BreconTile+MultiTexture.
Here is small test.
But what about render time? I made some test with MultiScatter today - simple scene with 50 copies of one tree scattered
1. by Multiscatter using Bitmap and CoronaBitmap.
2. by CoronaScatter using Bitmap and CoronaBitmap.
3. scattered manually using Bitmap and CoronaBitmap.
All these variants gave much longer render time with CoronaBitmap.
Last thing I've done I've replaced CoronaBitmap in opacity slot by regular Bitmap - render time is the same as in case of use only regular Bitmap everywhere.

All those renders are PT+UHD Cache. Tried PT+PT - the same results

BTW, how to put an image in between the rows, not only on the bottom of the post? :)

[romullus]

Just to rule obvious out - what interpolation did you use in CoronaBitmap?

[Bormax]

I used default Bilinear interpolation

[Bormax]

From my tests it's clear that using CoronaBitmap in opacity slot slow down rendering process. Using CoronaBitmap in all other slots doesn't speed up rendering. Now it's interesting for me - am I the only one who have this issue?

[romullus]

I didn't notice slowdown with opacity maps, but i haven't made much testing with it. As for general speed, i tend to agree, most of my scenes shows virtually no improvement with CoronaBitmap and few renders even slower. Although there was some cases where CoronaBitmap advatage was clear. And there's recently discovered bump issue... Well, i still undecided if i want to go fully with CoronaBitmap yet.

[hubrobin]

Hey,

We have made a change in CoronaBitmap that should fix this. Once the next build is out you can give it a go.

Cheers, Robin