Developer-ready AI PCs - out of box support for DirectML on NPU & GPU
summary
Shows a working Samsung app that uses DirectML on NPU & GPU to offload inference (demo catalogs a city photo; NPU utilization ticks up and search finds it), claims Core Ultra PCs can run up to 500 LLMs and Intel plans ~100 million Core Ultra devices in 18 months, and demonstrates one-line API offload from GPU to NPU, ONNX Runtime and WebNN preview support, FP16 numerics, and tooling like VTune. Includes a prompt‑injection reliability demo and says DirectML deployment is available on partner devices in retail today.
AI-written summary · about
chapters (10)
- 0:00Introduction to AI PCs
- 1:15Why now for AI PCs
- 3:23DirectML developer support
- 4:46Microsoft and Intel partnership
- 5:48Developers and Intel NPUs
- 7:32Security in AI development
- 8:38Demo: DirectML on Samsung AI PC
- 10:23Developer training next 24 months
- 11:27Key takeaways for developers
- 13:14Final thoughts on AI PC tools
related talks
50mLakshmi Narasimhan of Intel at Machine Learning Developers Summit 2019
28mDeveloper Essentials for the New Era of AI
12mPreview: ML Serving
50mML and AI at the edge with Google Coral, EdgeTPU and Cloud Platform
26mKeynote: Artificial Innovation Accelerates Artificial Intelligence
16mUnleashing the Future of AI Development
30mCloud? No Thanks! I’m Gonna Run GenAI on My AI PC
37mDeep Learning Neural Network Acceleration at the Edge