Global Edge AI Suite for Qualcomm Snapdragon

Build, compile, quantize, and deploy secure, low-latency AI models locally on Qualcomm Snapdragon NPUs and Oryon CPUs with Aura SDK, Aura Models, and Aura Compiler.

Target Platform: Qualcomm Snapdragon X

Qualcomm Snapdragon Edge AI Ecosystem

Bypass cloud hosting costs and latency. Ingest, compile, quantize, and run generative AI models locally on Qualcomm Snapdragon co-processors.

©Qualcomm

Qualcomm Snapdragon Edge AI Suite

The unified platform to compile custom models, run hardware-accelerated local inference, and manage your device fleet on Snapdragon X, automotive, and mobile chips.

Explore Our Edge AI Platform >

Aura SDK: Direct Hexagon NPU Access

Isolate inference inside the Hexagon NPU to optimize energy-efficiency and compute density. Leverage Hexagon Vector/Matrix extensions (HVX/HMX) for parallel fused multiply-add operations.

Explore Genie Engine >

Zero-Copy Memory Mapping

Directly map model weight tensors into ADSP address spaces, bypassing kernel validation and serialization delays for maximum generation speed.

Read Memory Design >

Aura Engine: ONNX Runtime CPU Fallback

Decouple model execution by routing non-supported operators and complex activation layers to the Qualcomm Oryon CPU cores, guaranteeing compatibility with standard ONNX graphs.

Explore CPU Engine >

Zero-Cost Rust Bindings

Interfacing with QNN C-APIs securely and without overhead. Enjoy Rust memory safety and zero-cost abstractions for production edge applications.

View API docs >

VBS Kernel Security Bypass

Interact directly with Hexagon registers. Disabling Virtualization-Based Security (VBS) allows unsigned user-space libraries to map model memory spaces with low overhead.

Setup VBS Guide >

Quantization and Precision Optimizations

Compile models using w4a16 (4-bit integer weights, 16-bit activations) for Hexagon tensor units, and int8/q4 uniform profiles for the Oryon CPU fallback cache hierarchy.

Explore Quantization >

Enterprise Edge AI Platform

Artificial Intelligence Lab delivers a complete, global software stack for localized AI. Build custom edge applications, compile raw PyTorch models for Snapdragon Hexagon NPUs, deploy optimized models securely, and gain deep hardware telemetry.

Get Aura SDK on GitHub Explore Platform Models >

©Qualcomm Snapdragon

Hardware-Level Performance & Safety

Our inference engines are compiled in native Rust to guarantee thread safety and memory isolation when executing models on Snapdragon X silicon.

Read Platform Overview >

NPU Compute Power (TOPS)

Dedicated AI compute across Snapdragon X processor models.

Supported Hardware & Runtime Dependencies

QAIRT v2.45

QNN SDK

Hexagon v73

Oryon CPU

Windows ARM64

Rust Native

Developers build faster with Artificial Intelligence Lab

Accelerating local inference. Learn how developers leverage Hexagon NPUs and our optimized Edge AI suite to run on-device models on Snapdragon X laptops.

Explore Our Edge AI Suite >

Qualcomm

Rust Native

ONNX Runtime

Windows ARM64

"With Aura SDK, our local AI assistant runs directly on the user's Copilot+ PC without sending sensitive telemetry to the cloud. The zero-copy NPU mapping cut activation latency in half."

Maximilien Grzeczka

CEO & Founder, ARL

View Project GitHub >

35 TPS

average generation throughput on Hexagon NPU (Llama 3 w4a16)

<10 ms

initial token activation warmup time with memory-mapped weights

"Interfacing with Qualcomm's QNN direct libraries was simplified through Aura's Rust abstractions. We get 30+ TPS on Phi-3 models using the Hexagon NPU, bypassing the cloud overhead."

Maximilien Grzeczka

CEO & Founder, ARL

View Benchmarks >

Developer Resources

Explore our deep-dives, research publications, and hardware integration guides. Learn how to leverage Qualcomm Snapdragon X NPU hardware natively with Rust to design next-generation edge applications.