llama.cpp Backend Documentation

Vulkan and SYCL Backends

The official llama.cpp documentation explicitly lists Vulkan and SYCL as supported backends. The backend table identifies SYCL specifically for Intel GPU targets. This documentation distinguishes these from other hardware-specific implementations. It does not claim universal compatibility across all devices. The presence of these backends indicates specific architectural support within the project's current build configuration.

Hybrid Inference Limitations

The documentation describes CPU+GPU hybrid inference as a method to partially accelerate models. It specifies this applies to models larger than total VRAM capacity. This mechanism allows partial offloading when memory constraints exist. The text does not guarantee full acceleration or specific performance metrics. It defines the operational scope for handling large model sizes within available hardware resources.