junkpile

08 — wgpu Backend Lab

A standalone Tauri 2 + Rust/wgpu example for inspecting which native GPU backends and adapters are available, then running the same WGSL workload through the selected backend.

What this example demonstrates

• A raw Tauri renderer window with no WebView in the render path.
• Adapter enumeration through Instance::enumerate_adapters.
• Backend selection before instance creation through WGPU_BACKEND.
• Optional adapter-name filtering through WGPU_ADAPTER_NAME.
• Surface compatibility, formats, present modes, and alpha modes.
• Hardware limits for textures, buffers, bind groups, and compute workgroups.
• Feature reporting for timestamp queries, subgroups, f16 shaders, and compressed textures.
• One identical WGSL workload for comparison across Metal, Vulkan, DX12, and the optional GL path.

Run

npm install
npm run dev

The default run allows wgpu to choose from its primary native backend set.

Explicit backend scripts

npm run dev:metal
npm run dev:vulkan
npm run dev:dx12
npm run dev:gl

The helper script sets WGPU_BACKEND in a cross-platform way before launching Tauri.

Select an adapter by name

macOS/Linux:

WGPU_ADAPTER_NAME="Apple M3" npm run dev

PowerShell:

$env:WGPU_ADAPTER_NAME="NVIDIA"
npm run dev

The match is case-insensitive and checks whether the adapter name contains the supplied text.

Platform expectations

Platform	Primary native backend	Other useful test paths
macOS	Metal	Vulkan through MoltenVK when available
Windows	DX12 or Vulkan	The other native backend for comparison
Linux	Vulkan	GL/GLES as a compatibility comparison

A single physical GPU may appear as multiple adapters when more than one backend exposes it.

macOS Vulkan note

This project enables wgpu’s vulkan-portability Cargo feature on macOS, but that does not install a Vulkan loader or MoltenVK. npm run dev:vulkan may report that no Vulkan adapter is available. Metal remains the native Apple path.

The lab reports exactly what wgpu can enumerate rather than claiming a backend exists because its code was compiled.

Fixed dependency line

This example intentionally matches the versions validated by examples 00–07:

Tauri       2.11.5
tauri-build 2.6.3
wgpu        29.0.4

Architecture

WGPU_BACKEND / WGPU_ADAPTER_NAME
                 ↓
          wgpu Instance
                 ↓
       enumerate GPU adapters
                 ↓
choose a surface-compatible adapter
                 ↓
      Device + Queue + Surface
                 ↓
       identical WGSL workload
                 ↓
 Metal · Vulkan · DX12 · optional GL

Important limitation

Backends cannot be changed after the wgpu instance, adapter, device, and surface have been created. Restart the application with a different launch script to perform a valid backend comparison.

This site is open source. Improve this page.