Tile-based level of detail for the parallel age.

Area of Science:

• Computer Graphics
• Parallel Computing
• High-Performance Computing

Background:

• Modern PCs utilize multiple Central Processing Units (CPUs) and Graphics Processing Units (GPUs) for enhanced performance.
• Clustering multiple PCs is a common strategy for achieving high-performance, interactive graphics.
• Efficient parallelization of rendering tasks with level of detail (LOD) is crucial for maximizing resource utilization.

Purpose of the Study:

• To present a novel approach for parallelizing rendering with level of detail using hierarchical, screen-space tiles.
• To efficiently parallelize the rendering workload across CPUs, GPUs, and multiple PCs.
• To achieve good resource utilization and adaptive load balancing in parallel graphics systems.

Main Methods:

• Implementation of a hierarchical, screen-space tiling strategy for rendering.
• Association of 'adapttiles', 'render tiles', and 'machine tiles' with CPUs, GPUs, and PCs, respectively.
• Utilizing adaptive tile sizes for load balancing and an LOD system for independent CPU/GPU load management.

Main Results:

• Demonstrated efficient parallelization of rendering tasks on both single PCs and clusters of PCs.
• Achieved good resource utilization through adaptive tile sizing and LOD management.
• Successfully balanced the workload across multiple processing units (CPUs and GPUs).

Conclusions:

• The proposed hierarchical tiling approach effectively parallelizes rendering with level of detail.
• The method enables efficient workload distribution and resource utilization in multi-CPU/GPU systems and PC clusters.
• Adaptive tile sizes and LOD management are key to balancing load on CPUs and GPUs.