<--back--      home      --next-->

This summer, we rebuilt it to work on a cluster of PCs. This volume renderer also features software compositing. The pipeline for both is essentially the same. (Right about here is where I got cut off in the interest of saving time for debate.) We were able to render the same sized volume with just about half the performance. Al McPherson presented these results last August at a Siggraph tutorial. We used 32 nodes for both rendering and compositing.

While the graphics cards we chose had a great deal of texture memory, they lacked the performance and special features of the latest generation. We know we can do better.