GPUDevice lost error: unresolved recovery strategy in WebGPU

Answer to the question

The WebGPU Level 1 specification does not mandate automatic adapter fallback after a GPUDevice.lost event. Recovery behavior is left to each browser implementation and to the application.

Confirmed facts

• The spec defines the GPUDevice.lost event but explicitly states that handling device loss is implementation‑defined.
• Current observations: Chrome attempts to reuse the same adapter when the GPU context can be reset; Firefox requires the developer to call navigator.gpu.requestAdapter() and requestDevice() again; Safari treats device loss as fatal and does not attempt any fallback.
• None of the major browsers expose a flag or API to choose between automatic fallback and manual recovery.

Steps to verify behavior in your environment

1. Create a minimal HTML page that initializes a WebGPU device.
2. Issue a known‑timeout compute shader (e.g., a loop with a large iteration count) to provoke GPUDevice.lost.
3. Attach a listener to GPUDevice.lost that logs GPUAdapter.description before the event and then calls navigator.gpu.requestAdapter() followed by requestDevice().
4. Run the page in Chrome, Firefox, and Safari (with WebGPU enabled) and compare the logged adapter descriptions to see whether the same adapter is reused.

Missing diagnostic detail that could change the recommendation

Whether the system exposes more than one GPU adapter (e.g., integrated vs. discrete) influences whether automatic fallback is even possible. If only a single adapter exists, fallback cannot occur regardless of browser policy.

Please confirm: Does your machine report multiple GPU adapters via navigator.gpu.requestAdapter()?