A Memory Efficient Encoding for Ray Tracing Large Unstructured Data.

Area of Science:

• Computer Graphics
• Scientific Visualization
• GPU Computing

Background:

• Modern GPUs offer potential for efficient unstructured data rendering.
• Large meshes require significant GPU memory for element representation and acceleration structures, posing a practical limitation.

Purpose of the Study:

• To develop a memory-optimized encoding for large unstructured meshes.
• To enable efficient rendering of complex datasets on GPUs by optimizing memory usage for mesh data and acceleration structures.

Main Methods:

• Described a novel memory-optimized encoding technique for unstructured meshes.
• Integrated encoding of both mesh data and sample reconstruction acceleration structures.
• Ensured fast random-access sampling capabilities for rendering.

Main Results:

• The encoding efficiently handles large unstructured meshes and their associated acceleration structures.
• Demonstrated rendering of a 2.9 billion element mesh on a single GPU.
• Showcased rendering of a 6.3 billion element mesh on a dual-GPU system.

Conclusions:

• The memory-optimized encoding overcomes GPU memory limitations for large unstructured meshes.
• Enables high-quality rendering of massive datasets on commodity GPUs.
• Paves the way for real-time visualization of complex scientific data.