← All resources

Optimizing spatiotemporal variance-guided filtering for modern GPU architectures

Auto-discovered

Paper Advanced

Rostyslav Pikulsky

Visit resource ↗

Description

Optimizing spatiotemporal variance-guided filtering for modern GPU architectures. Optimizing Spatiotemporal Variance-Guided Filtering for Modern GPU Architectures presents a Journal of Computer Graphics Techniques paper by Rostyslav Pikulsky on speeding up SVGF spatial filtering without changing filter output.

It analyzes Falcor SVGF stages, identifies a-trous wavelet iterations as the main cost, and proposes pixel processing layouts and bandwidth reductions for dilated convolutions on modern GPUs. Benchmarks on AMD Radeon RX 7800 XT and NVIDIA RTX 2080 Super report about 1.3 to 2.5 times faster runs across static and dynamic Full HD scenarios while preserving quality.

It is for advanced real-time path-tracing and denoising engineers who already know SVGF or similar wavelet reconstructors and can work in compute-shader full-screen passes.

The paper is free to read as a JCGT HTML/PDF article with timing tables and implementation discussion based on Falcor.

It is useful when SVGF already fits the quality budget but still consumes too much of a 60 Hz frame on current hardware. Link last verified accessible on 2026-09-29. Link last verified accessible on 2026-09-28.