Linear Model Analysis of Real-Time Rendering - Real-Time Rendering: Computer Graphics with Control Engineering - page 45

Step 2 was run again with another type of object and the same types of data were
collected. This process was repeated for all types of objects in the 3D scene to enable
us to derive the system models for rendering all types of objects separately.
Finally, the application was run with all types of objects displayed and data
sets collected by varying the object counts for all the objects. The purpose of this
step was modelling the full application so that the overall system model could be
compared with the sum of the individual system models obtained in steps 1 and 2.

3.8.2.1 Test Application
A test application was adopted from the NVIDIA DirectX 9.5 SDK. This sample
application demonstrated rendering using the hardware instancing technique. The
3D scene in this application consisted of three types of objects (rocks only, space-
ships only, and both rocks and spaceships as shown in Figure 3.18). The test applica-
tion allowed the user to switch off the rendering for any type of object and change
the number of objects (for each type) to be rendered as well.
Our data collection procedure started by setting the application to render only one
type of object (rocks). The triangle count (input) and frame rate (output) data pairs
were collected over multiple object counts within the allowed range. The application
then ran with only spaceships displayed. The same data pairs were then collected
for a range of object counts. Subsequently, the application was run with both types
of objects and the same data collection process. The intent was to constrain the
rendering process to specific types of objects so that we could perform black-box
modelling to develop the respective system models.

FIGURE 3.18

(See colour insert.) Screenshot of test application in superposition experiment.