WebGPU in PyNGL

Alongside the OpenGL stack, PyNGL ships a parallel renderer built on WebGPU via the wgpu Python package. WebGPU is the modern successor to OpenGL: instead of a hidden global state machine you describe your rendering up front as explicit pipeline objects (shaders, vertex layouts, and render state bundled together), then record draw calls into a command encoder each frame. Shaders are written in WGSL rather than GLSL.

Everything lives in the ncca.ngl.webgpu package, which exports four things:

  • WebGPUWidget — a PySide6 QWidget base class that hosts the renderer. It renders offscreen with wgpu into a colour buffer that the widget blits to the screen with QPainter, so there is no OpenGL context, canvas, or swapchain to manage. Subclass it and implement paintWebGPU() and resizeWebGPU(w, h).
  • PipelineFactory — a registry that creates ready-made render pipelines (points, lines, triangles, triangle strips, and instanced geometry), mirroring the OpenGL VAOFactory pattern. You can register your own pipeline classes without touching library code.
  • PipelineType — the enum of built-in pipeline types the factory knows about.
  • NGLToWebGPU — helpers that translate NGL type names into WebGPU vertex formats and strides (e.g. "vec3" → "float32x3").

All the maths and geometry classes are shared with the OpenGL stack — Vec3, Mat4, look_at, perspective, and PrimData work unchanged.

One thing to remember

WebGPU's clip-space depth runs from 0 to 1, whereas OpenGL's runs from −1 to 1. Always build projection matrices with perspective(..., PerspMode.WebGPU) — with the default (OpenGL) mode your geometry will clip or z-fight.

The pages in this section

  1. Getting Started with WebGPU — the WebGPUWidget lifecycle and a minimal working application.
  2. The Built-in Pipelines — a tour of all fourteen PipelineTypes and the set_data / update_uniforms / render contract, following the bundled demo app.
  3. Custom Pipelines — writing your own WGSL shaders with CustomShaderPipeline, or subclassing BaseWebGPUPipeline and registering it with the factory.
  4. WebGPU API Reference — full auto-generated documentation for every class in the package.

Try it right now

The package ships a demo application that cycles through every built-in pipeline. If you have the library installed you can run it immediately:

uv run python -m ncca.ngl.webgpu

Use the Left / Right arrow keys to switch pipelines manually, Space to pause the animation, A to toggle automatic switching, and Escape to quit. Its source (src/ncca/ngl/webgpu/__main__.py) is the reference example for the built-in pipelines and is walked through in The Built-in Pipelines.

For larger complete programs — lit meshes, textures, shadows, compute — see the WebGPU examples in PyNGLDemos.