Overview of Qualcomm Snapdragon Neural Processing Engine
summary
Runtime for Snapdragon SoCs that loads DLC deep learning containers, runs whole networks on CPU, GPU or Hexagon DSP, and provides converters from Caffe (ONNX import imminent). SDK v1.13.0 ships snappy_net_run, snappy_diag and snappy_bench, Python/C++/Java APIs, and x86 conversion tools. DSP backend natively uses 8-bit quantized models while CPU/GPU use 32-bit with float16 and 16/32 hybrid modes; user-defined layers let unsupported ops run on CPU. Supports 845, many 600-series, 450, 820 AGL and Snapdragon Flight.
AI-written summary · about
chapters (8)
- 0:00Introduction to SNPE
- 1:05SDK components overview
- 3:39Runtime and hardware support
- 8:32Model conversion workflow
- 10:01User-defined layers and input formats
- 11:06Quantization and backend choices
- 13:55Developer resources and examples
- 14:15Q&A session
related talks
16mRealize your vision with a guide to building Windows on Snapdragon
29mQualcomm® AI Stack for developers and extension to on-device AI
15mBuilding Gen-AI experiences using LLM and LVM
32mDeploy AI with Simulink
39mQualcomm Technologies and the power of Windows AI Q&A
05mDrive UX innovation on Windows devices powered by Snapdragon
54mHarnessing On-Device AI to Transform the Smartphone Experience
14mAccelerating On-Device AI Development