Graphics: from a 100x data expansion to a 100x compression function

In the mid 1980s, a typical large data visualization, involved the input of geometry in the order of 10,000 triangles, for rendering into about a million display pixels. However, today, we see input data sets growing to a billion triangles, while the generated display has only grown to about 10 million pixels. Consequently, large data visualization have, over the years, changed from a data expansion, to a data compression function.This four orders of magnitude change has evoked renewed interests in ray tracing rendering techniques, of which its performance can be made more input geometry-quantity insensitive. This is as opposed to scanline rendering techniques, such as OpenGL, that can be made more output pixel-quantity insensitive.For those staying with scanline techniques, the enormous growth in geometry count has made it necessary for the development of scalable parallel rendering. This is where tens of Graphics Processing Units (GPUs) are coordinated to render one large geometric data set. After the pixels are produced by each of these GPUs, ways of compositing their individual output into a single display, with feedback loops for dynamic load balancing, may be necessary.As the crossover "from expansion to compression" continues, it will become increasingly practical, in certain remote visualization sessions, to invent ways to transmit only the generated pixels; instead of the traditional method of transmitting the entire geometric data set for pixel generation at the remote user's station. Add to this the advances in display, lighting and input technologies for mobile handheld devices, interesting new applications may evolve for the scientific, engineering and creative users.