Therefore, when dynamically balancing tasks (or load-balancing) between different rendering engines, the above approach may require special care to adjust the dispatched tasks by communicating task-progress information back to a central dispatch (e.g., dispatch 120 of FIG. 1) and then configuring the central dispatch to adjust the sub-region(s) assigned to each rendering engine, in order to achieve a better load balancing. Alternatively, the central dispatch may be configured to subdivide an image into sufficiently small tiles such that a round-robin type of tile assignment can hide (or mask) an uneven load distribution because the variance in time to compute a given tile becomes sufficiently small. However, this approach increases the overhead that is performed for each rendering task, adds additional inter-process communication, and/or decreases the efficiency of parallelism. In addition, this approach requires exclusive task dispatch logic to add/remove regions to/from each rendering process.

In contrast, according to various embodiments of the present disclosure, communication between the rendering engines is not required during rendering, and thus parallel performance is not impacted.

Aspects of the present disclosure are directed to a method of distributing tasks (e.g., sampling locations within a particular pixel) to each of multiple rendering engines that are used to render an image. For example, each rendering engine may operate (or reside) at a respective computer (or host) of a plurality of computers. Aspects of the present disclosure are directed to addressing one or more of the following areas.