-
Notifications
You must be signed in to change notification settings - Fork 0
GPU Selection and Vulkan
Recent Neo builds include explicit GPU-selection support and a Vulkan processing path for supported operations.
On multi-GPU systems, automatic selection can occasionally choose an integrated GPU when a discrete GPU would be preferable. Explicit selection allows Neo to target the intended GPU.
This is particularly useful for systems with:
- Integrated GPU + discrete GPU
- More than one discrete GPU
- Mixed-vendor GPUs
When a build supports it, Neo attempts to keep compatible GLSL / ONNX processing on the selected GPU rather than unexpectedly splitting work across adapters.
Backend limitations can still apply. For example, TensorRT is NVIDIA-specific.
Vulkan is used by Neo as an alternative GPU processing path where supported. Performance differences depend on the GPU, driver, filter chain, and data-transfer path.
If the default path is stable and fast on your system, changing graphics API only for theoretical performance is unnecessary.
Depending on the current build and backend state, some GPU/backend changes may require resources to be recreated. Follow the UI behavior of the version you are using.