Shreekanth Pavar wrote:

To put the approach taken by Manta into context, would you agree that it uses a
"Sort-Last" approach to parallel rendering as defined in the paper below ?

A sorting classification for parallel rendering
S. Molnar, M. Cox, D. Ellsworth, and H. Fuchs.

In a sort last renderer the geometry is divided between parallel processors and independently rasterized, then fragments are sent to compositors which resolve visibility. The compositor performs the sorting step after rasterization which might have produced more than one fragment for a given pixel on different processors.

In Manta, visibility is computed by shooting a ray through the geometry and finding the first intersection. There isn't a separate composite step which performs additional sorting on "fragments" (in this context really shaded intersection points). Using a shared memory system we avoid the need to perform any type of external geometry or fragment sorting since any processor can access the entire geometry.


D.E. DeMarle, C.P. Gribble, S. Boulos, S.G. Parker. “Memory Sharing for Interactive Ray Tracing on Clusters,” In Parallel Computing, Vol. 31, No. 2, pp. 221--242. 2005.

Abe