Blender internal VS Cycles on a simple scene

I was bored so decided to render a simple scene on both BI and Cycles and see what the results would be. I tried to adjust the settings so that the scenes would match as well as possible, this includes the number of samples used to get about equal noise levels. Susane is pure matte, the cube is matte with 0.3 amount of sharp glossy, the sphere is pure sharp glossy and the plane is also pure default white matte. For the lighting there are two point lamps, another has light yellow and the other light blue tint.

First: 0 bounces (except reflection/glossy depth set to 2)
It’s amazing how the images look exactly alike.

BI 17sec
http://www.pictureshack.us/thumbs/71102_SimpletestscnBI_17sec.jpg

Cycles 29sec
http://www.pictureshack.us/thumbs/48405_SimpletestscnCycles_29sec.jpg

Next: again 0 bounces but now with ambient occlusion
Still not much of a difference.

BI 39sec
http://www.pictureshack.us/thumbs/11773_SimpletestscnAOBI_39sec.png

Cycles 48sec
http://www.pictureshack.us/thumbs/75828_SimpletestscnAOCycles_48sec.jpg

And then: 3 bounces
Difference is obvious, it’s impossible to get BI look like Cycles without real ray tracing for the bounces. Also while BI doesn’t really produce any noise it’s a lot less accurate. I only upped the number of diffuse samples in Cycles a bit, but there’s still some noise.

BI 60sec
http://www.pictureshack.us/thumbs/66066_SimpletestscnBI3bounces_60sec.jpg

Cycles 73sec
http://www.pictureshack.us/thumbs/3057_SimpletestscnCycles3bounces_73sec.jpg

The result…? Well looks like BI can be a bit faster on some very basic scene this time, but anything more than that, you’ll probably want to use Cycles.

Build used for both BI and Cycles was mingw64 2.64
Cheers!

By setting the bounces to zero in cycles you are telling the render to basically function just like blender internal, which uses only direct lighting without any bounces.

Keep in mind that point lamps are not importance sampled in Cycles yet. I imagine the render times will drop when this functionality is added.

That was the idea yes. I was interested in how Cycles’ speed would compare to the old engine on a same scene and settings.
But I was surprised that you can actually render basically the same image on both BI and Cycles, I thought there would always be a difference.

Btw I used SIMD QBHV on BI which was the fastest mode.

I discovered that optimal tile size for BI (at least this scene) is 96x96. Changing tile size from 16 to 96 reduced the rendering time of the first scene from little over 17sec to little under 8sec. Looks like BI performs better on a little larger tile size. The speed was about same on both “normal” and mingw build.

Different tile sizes good for different scenes. it all depends on what is going on in your scene (lots of sky? lots of hair/grass? lots of transparencies? reflections? etc.etc.) & how big your final render size is.

how set off ambient occlusion in cycles?

The best way to do this is not to set it ON :slight_smile:
Look among the “World” properties.

I decided NEWER EVER use BI for more complex scene - it is slow, and extremely slow with glossy surfaces
Much faster is Yafaray/Cycles with AO
Mitsuba with GI is also very fast (faster than Cycles)