Six Ways to Draw Vangers with WebGPU: Real-Time Rendering of Editable Multi-Layer Height Fields

A new rendering study takes on one of the messier terrain problems in games: Vangers-style maps with two solid intervals per sample, local destruction, and no room for a static preprocessing pass. The work compares six methods in a single WebGPU engine using the same data path and WGSL shaders, so the tradeoffs are about representation and runtime behavior rather than API differences.

The six techniques are height-field ray marching, voxel-accelerated ray marching, sliced proxy geometry, per-sample bar rasterization, compute scattering, and a fitted triangle mesh. At the camera angle used in the original game, they can look similar. At eye level, the differences show up fast: point scattering can miss coverage, slicing can band, and an oversimplified mesh can fail to preserve walls and other terrain features.

On the measured hardware, a greedy triangulated irregular network had the lowest mean frame time across devices, but the fit cost was driven by the second layer rather than simple floor relief. Keeping that mesh editable is expensive, though: 319 MiB of GPU geometry and 535 MiB of CPU triangulation are retained. That makes the result especially relevant for teams balancing runtime speed against live terrain editing, destruction, and memory budgets.

The engine, benchmark harness, and a one-command measurement setup are being released alongside the work, which should make it easier for other teams to reproduce the numbers or swap in their own terrain data. For developers shipping deformable worlds, the practical takeaway is that the fastest representation may not be...