Tutorial
Graphics Performance

Background

Graphics performance is often measured by framerate, in frames per second, usually abbreviated as fps but more scientifically correct as f/s. Frametime is the inverse of framerate, and is usually measured in milliseconds per frame, i.e. ms/f.

Whilst framerate is an intuitive and important measure, and relates directly to the user's real-time experience, it has the property of reducing as the processing time increases, whereas frametime increases.

Another aspect of timing is that when using doubling/halving of tesselation (or other parameters) need to be careful about having a non-linear scale. A doubling of tesselation in each direction gives a four-fold increase in the number of polygons.

Use the attached spreadsheet, which has measured times from a program. Use libreoffice (which should open automatically). Use a 'scatter graph' and cntl to select different columns

Exercises:

Plot the following graphs:

1. Number of triangles versus frametime
2. Number of triangles versus framerate
3. Grid tesselation versus frametime
4. Grid tesselation versus framerate

Which graph shows the performance relationship between the variables most clearly? What is that relationship? What do these result say about the bottleneck?

Further exercises:

1. Run the fmbm.c program (see assignment 1), or your own program and measure and plot times. Just use immediate mode.
2. Enable vsync using nvidia-settings. Run a test. Are the results as expected?
3. Investigate whether enabling running multiple instances of your program similtaneously changes results. The sutherland machines have quad core CPUs (Intel i7s).