Anti-aliasing

As a hybrid engine, REDsdk offers several anti-aliasing methods. This page reviews all the possible controls an application may have over using anti-aliasing in a REDsdk application.

Anti-aliasing on the GPU

REDsdk has three main anti-aliasing techniques that can be used on the GPU:

Method	Activation	Quality	Performance	Comments
Hardware AA	RED::WindowRenderInfo::SetHardwareAntiAlias	Average.	Good.	May consume a lot of memory. Not available on offscreen VRLs. Quality may vary with hardware generation.
Engine AA no superscale	RED::IViewpointRenderList::SetEngineAntiAlias	Good.	Average.	Do not consume any memory.
Engine AA with superscale	RED::IViewpointRenderList::SetEngineAntiAlias	Good.	Good.	Consumes memory in relation with the AA value.
FXAA	RED::RenderShaderAAFilter	Low.	Very good.	Is a post-processing effect, not a true anti-aliasing, so it will never capture originally missed details.

Anti-aliasing on the CPU

Method	Activation	Quality	Performance	Comments
Software AA	RED::IViewpointRenderList::SetSoftAntiAlias	Good.	Good.	Has a significant effect on render time.

On the example scene below:

The example scene used to measure software AA performances

We get the following numbers, on a standard Core i7 computer, rendering a 1000 x 600 pixels image with average quality settings:

AA value	Render time
x 1	1 mn 15 sec.
x 4 (= 16 AA samples per pixel)	2 mn 24 sec.
x 8 (= 64 AA samples per pixel)	6 mn 20 sec.
x 16 (= 256 AA samples per pixel)	19 mn 10 sec.

We can see that increasing the AA value does not increase the render time linearly: on multiplying the AA value by x4, we get roughly two to three times the render time. This is important to be aware of, as using a high AA value is often a must have for high quality images.

Note that increasing the image quality accordingly will reduce the AA overhead ratio, as primary rays will account for less in the rays being propagated to compute the image.

Modifying engine data