<--back--      home      --next-->

The catch is...
These cards weren't designed with visualization and its needs in mind.

Some features which we rely on, such as 3D texture, just haven't been given the attention that they could have been given.

As I mentioned before, PC interconnects are less than optimal for our high throughput applications. Things like frame buffer read-back are lagging way behind what we are use to from the "Big Iron".

This is especially bad for large scale visualization. Shown here is an example parallel graphics pipeline. Data resides on disk somewhere and is read into main memory. It is then processed and sent to the graphics hardware for rendering. The intermediate frames are then read back to main memory and composited for display. The red box indicates the graphics hardware's position in this process.