You copied the Doc URL to your clipboard.

3.2.4. Bottlenecks move between processors

This section describes how bottlenecks move between processors and the profile of an ideal application. It contains the following sections:

How bottlenecks move between processors

The performance bottleneck in an application can move between the different processors as optimizations are applied. The readings from Performance Analysis Tool can tell you where the bottleneck is likely to move to and if a processor is under-used.

Figure 3.3 shows a bar graph of frame rates for different parts of a system running an application.

Figure 3.3. Frame rate limitations of different system elements

Figure 3.3. Frame rate limitations of different
system elements

Comparing the bars indicates the following:

  • If you optimize the performance bottleneck, the next bottleneck is the component with second lowest bar. In this case the bottleneck is the CPU and the next bottleneck is the pixel processor.

  • These graph of the geometry processor has a much higher value than the overall GPU Frame Rate. This indicates that the bottleneck is not in the geometry processor. The large difference also indicates it is under-used.

    If a processor has spare processing capacity, consider if there are any processing operations that you can move to it. For example, you might be able to move operations from the CPU or pixel processor to the geometry processor.

Ideal application profile

An ideal application is limited approximately equally by all elements. A bar graph such as Figure 3.4 indicates the application is making good use of all elements.

In this case a single optimization is not likely to make a large impact on performance and you require multiple optimizations give a higher and more stable frame rate.

Figure 3.4. Ideal application equally limited

Figure 3.4. Ideal application equally limited

Was this page helpful? Yes No