Graphics Reference
In-Depth Information
Measured and Simulated Output
105
100
95
Simulated output
Measured output
90
5600
5800
6000
6200
6400
6600
6800
7000
7200
Frame
Ve rtex Count (Input)
×10 5
2
1.8
1.6
1.4
1.2
5600
5800
6000
6200
6400
6600
6800
7000
7200
Frame
FIGURE 3.8
Measured and simulated output of rendering application in Experiment 1.
For an ideal linear system, the hardware's processing capability should remain
unchanged at all operation ranges. This means that the time taken to process every
vertex should be constant. However, we can see in Figure 3.7 a non-linear trend in
vertex processing time as vertex count increases. The relationship may be approxi-
mated with multi-linear segments that fit the curve shown in the same figure. With
this prior knowledge of the model, we proceed with model identification based on
data measured in Experiment 1.
The simulated output of the derived model is shown in Figure 3.8. Note that the
derived model produced reasonably accurate results in comparison to measured data
with a maximum error less than 5 FPS in steady-state and best fit value of 84%. The
best fit computation is:
ˆ
=−
|
yy
yy
|
Best Fit
 
 
1
(3.13)
|
1
where y is the measured output, y is the simulated output, and y is the mean of
y . A 100% value corresponds to a perfect fit. This result validates our hypoth-
esis that the range is approximately linear. The model parameters are estimated
in MATLAB using the system identification toolbox; the values are provided in
Table 3.2.
In the second part of Experiment 1, we modelled the rendering of computer game
software as shown in Figure 3.4. Figure 3.9 shows the measured and simulated out-
puts of the system. It is noteworthy that the environment to be modelled becomes
 
Search WWH ::




Custom Search