Forge Lab — Experimental Realtime & WebGPU Prototypes

SYS.05 // R&D LABORATORY

FORGE LAB.

EXPERIMENTAL GRAPHICS · PARALLEL NETCODE · SPATIAL ALGORITHMS

BUILD_25.04.1BENCHMARK
WEBGPU_COMPUTE_BUFFER // DISPATCH_01
DISPATCH TIME: 0.86 MS
WEBGPU·2025.Q1

Compute Particle Vector Fields

Stress test evaluating 1.5 million particles simulated in WebGPU compute shaders. Calculates 3D curl noise and velocity advection fields in under 0.86ms per frame.

WebGPUWGSLCompute BuffersCurl Noise
BUILD_24.11.8ACTIVE
DUAL_CONTOURING // OCTREE_MESH
CHUNK EXTRACTION: 3.2 MS
PROCEDURAL·2024.Q4

Dual-Contouring Voxel Extractor

A multi-threaded Web Worker implementation of the Dual Contouring algorithm. Preserves sharp mechanical creases and geometric features while generating continuous isosurfaces.

TypeScriptWeb WorkersDual ContouringWASM
BUILD_25.02.3PROTOTYPE
FOVEAL_EYE_VECTOR // GAZE_TRACK
SHADING LOAD: -44%
XR·2025.Q1

Dynamic Foveated Multiview Pass

Stereoscopic single-pass multiview rendering pipeline integrated with eye-tracking gaze vectors. Concentrates GPU shading power in the foveal region for high-density spatial displays.

WebXRVisionOSOpenXRGLSL Shaders
BUILD_24.09.2BENCHMARK
HERMITE_QUATERNION // REWIND_CURVE
DESYNC DRIFT: < 0.02°
MULTIPLAYER·2024.Q3

Hermite Quaternion Extrapolation Netcode

Authoritative network state extrapolation evaluating Hermite quaternion interpolation. Eliminates visual stutter and rotational snapping during simulated 80ms packet bursts.

WebRTCTypeScriptQuaternionsNetcode
BUILD_25.01.5ACTIVE
RAYCAST_SUSPENSION // SIMD_128
STEP LATENCY: 0.38 MS
PHYSICS·2025.Q1

SIMD Raycast Suspension Solver

A deterministic 4-wheel vehicle suspension solver compiled from C++ to WebAssembly with 128-bit SIMD vector intrinsics. Resolves multi-surface friction in under 0.4ms.

C++WebAssemblySIMDRaycasting
BUILD_24.12.1BENCHMARK
CLUSTERED_AABB // 512_LIGHT_CULL
MAX LIGHTS: 512 ACTIVE
REALTIME 3D·2024.Q4

Clustered Forward+ Light Culler

GPU compute shader that slices the camera frustum into a 3D AABB grid (16x16x24 tiles) to evaluate up to 512 dynamic point lights with minimal draw overhead.

WebGL 2.0WebGPUClustered ShadingGLSL
BUILD_25.05.0RESEARCH
R&D PROTOTYPENEURAL_MESH_DISTILLATION // SPEC_INTAKE
STATUS: R&D INTAKE
AI EXPERIMENTS·2025.Q2

Neural Radiance Field Mesh Distillation

[R&D PROTOTYPE] Investigating automated pipeline to bake 3D Gaussian splat radiance fields into compressed Draco polygonal meshes with baked spherical harmonics.

PyTorchGaussian SplattingDracoWebGPU
[ INGESTION PROTOCOL ]
LAB STATE: 07 EXPERIMENTS COMPILED · CONTINUOUS R&D
STABILITY: ZERO UNHANDLED REJECTIONS