Yocto

The Genio Yocto Project provides a flexible environment for evaluating AI workloads using open-source frameworks. This section focuses on benchmarking tools and demonstration applications specifically designed for Yocto-based Genio platforms.

While this page highlights Yocto-specific resources, the core development tools, host-side SDKs, and foundational concepts are common across all Genio operating systems.

Software Architecture

Genio Yocto images integrate two primary open-source inference frameworks: TFLite and ONNX Runtime. These frameworks are built from upstream open-source projects and pre-packaged in the Board Support Package (BSP) to enable immediate evaluation.

The following figure illustrates the Yocto-specific AI software stack, focusing on the analytical AI paths.

AI software stack overview for Genio Yocto platforms

Supporting Scope on Yocto

The available AI software stack depends on the hardware capabilities of each Genio platform. The following table summarizes the support for TFLite (Online and Offline) and ONNX Runtime on Yocto.

AI Supporting Scope on Yocto

Platform

TFLite - Analytical AI (Online)

TFLite - Analytical AI (Offline)

TFLite - Generative AI

ONNX Runtime - Analytical AI

Genio 520/720

CPU + GPU + NPU

NPU

X (ETA: 2026/Q2)

CPU + NPU

Genio 510/700

CPU + GPU + NPU

NPU

X

CPU

Genio 1200

CPU + GPU + NPU

NPU

X

CPU

Genio 350

CPU + GPU

X

X

CPU

Note

Currently, Genio Yocto images do not support Generative AI (GAI) workloads. MediaTek plans to introduce GAI support for the Yocto Project in 2026 Q1. Developers requiring GAI capabilities should currently refer to the Android platform resources.