GPU Boids Spatial Hashing & Radix Sort Lab

FPS

58.5

Search Time

1.42 ms

Avg Neighbors

14.8

Atomic Contention

12 / frame

Presets Architecture Scenarios

Live Compute Telemetry

Active Algorithm	Atomic Grid O(N)
Total Active Boids	1024
Neighbor Search Latency	1.42 ms
Effective Frame Rate	58.5 FPS
Avg Flock Neighbors	14.8
Atomic Collisions (Interlocked)	12
Memory Bandwidth Traffic	4.2 GB/s
Pipeline Health Status	Balanced (Low Atomic Contention)

GPU Radix Sort Hashing Steps WGSL Simulation

Sorts particle Morton codes/cell IDs without atomics. Step 4-bit radix passes:

128B0

142B1

119B2

135B3

120B4

131B5

118B6

131B7

Simulation Physics & Weights

Swarm Size (N) 1024

Spatial Grid Resolution 32³

Separation Weight 1.5

Alignment Weight 1.0

Cohesion Weight 1.0

Architecture Note (Dawn & WebGPU): In WebGPU compute shaders, simple atomicAdd(&gridHead[cell], 1) works for modest flock sizes, but in dense flocks hundreds of threads serialize on the same memory address causing warp stalls. Radix sorting cell hashes eliminates write conflicts completely, creating contiguous memory layout for high cache hits.