Qualcomm AI Engine Direct allows clean separation in the software for different hardware cores. The SDK lets developers treat Qualcomm AI Engine Direct as a hardware abstraction API and port applications among those hardware cores. The API offers the right level of abstraction, handling capabilities such as graph optimization internally while leaving broader functionality like model parsing and network partitioning to higher-level frameworks.

Developers can regulate the trade-off between capabilities provided by the core-specific, back-end libraries and the cost in library size and memory utilization. The resulting compilation is a high-performance, nimble executable that occupies a minimal memory footprint while providing the highest performance.

The architecture is designed to provide modular, extensible, accelerator-specific libraries that form a reusable basis for full-stack AI solutions. It is used by Qualcomm Technologies, Inc. (QTI) within the Qualcomm^® Neural Processing SDK and as a delegator for third-party frameworks such as TensorFlow Lite and the ONNX runtime.

The Qualcomm AI Engine Direct architecture provides the main components – tools, device, backend, context, graph, operation package registry – for building, optimizing and executing network models on the best-suited core.

The Qualcomm AI Engine Direct SDK is available for download now. See how you can use its unified API and core-specific libraries for running full-stack AI solutions on Snapdragon^® processors. And find out more about the entire Qualcomm^® AI Stack.