Building Gen-AI experiences using LLM and LVM
summary
Qualcomm shows Snapdragon X platforms running generative models up to 13+ billion parameters on-device using a 45 TOPS NPU, claiming parity with cloud for some LLM/LVM tasks and 3.5X UL Procyon performance versus a leading AI‑PC. Tooling includes Qualcomm Neural Processing SDK, AI Engine Direct, AIMET quantization, ONNX Runtime QNN execution provider and DirectML offloads, Jupyter notebooks, profilers and quantization/accuracy checkers; examples include Stable Diffusion and live captions running locally.
AI-written summary · about
chapters (8)
- 0:00Snapdragon X AI platforms
- 1:13On-device AI performance
- 4:04NPU architecture deep dive
- 5:10AI development tools & SDKs
- 7:22Microsoft AI collaboration
- 9:11On-device Gen-AI models
- 11:19Developer resources & hybrid AI
- 12:44Snapdragon X Microsoft AI features
related talks
29mQualcomm® AI Stack for developers and extension to on-device AI
39mQualcomm Technologies and the power of Windows AI Q&A
14mAccelerating On-Device AI Development
08mUnlocking AI at Scale for PyTorch Developers
54mHarnessing On-Device AI to Transform the Smartphone Experience
37mHow to Optimize, Validate and Deploy ML Models On Device
26mHow to use Qualcomm AI Hub
17mOverview of Qualcomm Snapdragon Neural Processing Engine