I have a chapter in the GPU Pro book (edited by Wolfgang Engel) covering virtual texturing with CUDA. The chapter describes how CUDA can be used to accelerate the CPU-side processing pipeline for virtual texturing — specifically the feedback buffer analysis, tile prioritisation, and DXT compression of streamed tiles.
GPU Pro Chapter
The chapter walks through:
- The classic virtual texturing pipeline (feedback → decode → upload → resolve)
- Where the CPU bottlenecks occur in a software implementation
- Using CUDA to parallelise feedback analysis and tile decompression
- Benchmarks comparing CPU vs. CUDA implementations
GPU Pro is part of the well-known ShaderX / GPU Pro / GPU Zen series of graphics programming books. Having a chapter in it was a nice way to get the research out to a wider audience.
GTC Poster
I also presented a poster on the virtual texturing work at GTC (GPU Technology Conference) in 2009. The poster covered the academic research angle of the project — the work that would eventually feed into my PhD thesis at Ghent University.
The research showed that CUDA offloading could reduce per-frame CPU time for tile management by a significant margin, making it more practical to support very high-resolution virtual texture setups in real time.