Skip to content

Latest commit

 

History

History
60 lines (39 loc) · 2.64 KB

README.md

File metadata and controls

60 lines (39 loc) · 2.64 KB

Reference Starting Point: WebGL Forward+ and Clustered Deferred Shading

======================

University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 4

  • Maya Huizar
  • Tested on: Google Chrome on MacOS Sonoma, M2

Live Demo

Try it yourself! No installation needed!

Demo Video/GIF

Performance Analysis

This assignment has a considerable amount of performance analysis compared to implementation work. Complete the implementation early to leave time for analysis!

Milliseconds per Frame vs. Light Count

Milliseconds per Frame vs. Light Count

Settings:

  • Max Lights per Cluster: 500
  • Clusters Count: 16,384

This graph compares the performance of Naive, Forward+ (Clustered), and Clustered Deferred shading methods across different light counts. Render times are measured in milliseconds per frame (ms). As the number of lights increases, Forward+ and Clustered Deferred show significant performance advantages over the naive approach.

Milliseconds per Frame vs. Cluster Size

Milliseconds per Frame vs. Cluster Size

Settings:

  • Max Lights per Cluster: 500
  • Total Lights: 1,000

This graph explores how cluster size impacts the performance of the Forward+ (Clustered) method. Varying the cluster size helps to identify optimal configurations for different rendering scenarios. Render times are provided in milliseconds per frame (ms), showing the balance between granularity and performance.

Benefits, Trade-offs, and Optimization Insights

Is one of them faster?

  • Forward+ (Clustered) generally performs better for moderate to high light counts, offering a good balance of performance and scalability.
  • Clustered Deferred excels with complex scenes containing many lights, efficiently managing shading tasks through the G-buffer.

Trade-offs

  • Forward+: Effective for dynamic lighting, but can incur overhead with high light counts and poorly distributed clusters.
  • Clustered Deferred: Reduces redundant calculations but requires more memory, which may impact performance in bandwidth-constrained scenarios.

Optimization Notes

  • Ensure no unnecessary copies of large structs/arrays in shaders. Use pointers in WGSL to manage this.
  • Chrome's profiling tools are valuable for detecting bottlenecks in TypeScript and WGSL code.

Credits