Testing the implementation by scaling up
All features and the user interface are identical to Chapter 1. But our changes can be made visible by adding more and more instances. If you add the same 1,600 instances as in Figure 2.1, you will see much smaller matrix generation times. The values may be similar to Figure 2.5:
Figure 2.5: The compute shader version with 1,600 instances
The time for virtually the same matrix operations went down from ~24 milliseconds on the CPU to less than 6 milliseconds by using compute shaders. We won around 18 milliseconds of CPU time in every single frame!
Now let us add more models – many models. Let’s say we add a total of 4,000 instances of the example model. The resulting matrix generation times on your machine may be similar to the number in Figure 2.6:
Figure 2.6: The compute shader version with 4,000 instances
Even with 2.5 times the number of instances, the average matrix generation time of the compute...
