Cycles GPU device selection fallback behavior in Blender

Cycles Render Devices Backend Switching

Blender's Cycles renderer supports multiple compute backends including CUDA, OptiX (NVIDIA), HIP ROCm (AMD), Metal (macOS), and CPU fallback. These devices are configurable in Edit > Preferences > System > Cycles Render Devices, where users can enable multiple devices simultaneously with GPU prioritized over CPU.

The critical question is how Blender handles automatic fallback when a GPU kernel fails to compile or encounters out-of-memory conditions. While the documentation states that CPU fallback occurs, the exact trigger conditions and whether Blender automatically retries with alternative GPU backends remain unclear.

Device availability detection happens at startup and when preferences change, but the silent nature of fallback transitions means users may not receive clear feedback about which backend is active during rendering. This becomes particularly complex across different Blender versions where backend support and naming conventions shift.

The unresolved decision centers on whether the automatic retry mechanism should attempt alternative GPU backends before falling back to CPU, and how the UI distinguishes between partial backend failures versus complete fallback scenarios.

Key questions:

• Should Blender automatically retry with a different GPU backend on first kernel compilation failure, or require explicit manual device reselection?
• How can users reliably detect and distinguish partial GPU failures from complete CPU fallback in the UI and console output?
• What specific conditions in the system console or render logs definitively indicate a backend switch occurred?