AMD Ryzen AI NPU
Vendor: AMD
Category: NPU Neural Processor
Architecture: XDNA 2
Introduction
AMD Ryzen AI is an NPU integrated into AMD Ryzen 7040/8040/9000 series processors. Based on XDNA architecture, it supports ONNX Runtime and DirectML accelerated inference, suitable for AI applications on thin-and-light laptops.
Specifications
| Model | Compute | Memory | Interface | TDP | Process |
|---|---|---|---|---|---|
| Ryzen AI 300 (Strix) | 50 TOPS (INT8) | LPDDR5X (shared) | Integrated SoC | 15-54W | 4nm |
| Ryzen AI 200 | 16 TOPS (INT8) | LPDDR5X (shared) | Integrated SoC | 15-28W | 4nm |
Official Website
Driver Downloads
Windows
Linux
Related Documentation
OS Support
| Windows | Linux | macOS | Android |
|---|---|---|---|
| ✅ | ⚠️ (Experimental) | ❌ | ❌ |
Version History
| Version | Release Date | Description |
|---|---|---|
| Ryzen AI Software 1.2 | 2024 | NPU Studio graphical tool |
Performance Benchmarks
| Model | Task | Performance Metric |
|---|---|---|
| Ryzen AI 300 (Strix) | Windows Studio Effects | Real-time background blur/eye contact |
| Ryzen AI 300 | NPU TOPS | 50 TOPS |
| Ryzen AI 300 | Stable Diffusion (local) | ~30s/img |
Pricing
| Model | Reference Price | Notes |
|---|---|---|
| Ryzen AI 300 (Strix) | Provided with laptop | AI PC laptop |
| Ryzen AI 200 | Provided with laptop | First-gen AI PC |
Quick Installation
Windows 11
Ryzen AI NPU is automatically activated through Windows 11 drivers and ONNX Runtime. No additional installation required. Ensure system is updated.
# Verify NPU availability (via DirectML)
python -c "import onnxruntime as ort; print(ort.get_available_providers())"
Code Examples
Python (ONNX Runtime NPU)
import onnxruntime as ort
# Use DirectML / NPU backend
session = ort.InferenceSession("model.onnx", providers=['DmlExecutionProvider'])
results = session.run(None, {'input': input_data})
Architecture Highlights
- XDNA 2 Architecture: AMD second-gen NPU architecture, 50 TOPS meets Copilot+ PC requirements
- AI PC Positioning: Oriented toward daily AI assistant tasks (background blur, voice enhancement, text generation)
- CPU+NPU+GPU Co-processing: Three compute units working together, NPU handling low-power always-on AI tasks
Model Compatibility
| Model/Framework | Support | Notes |
|---|---|---|
| ONNX Runtime | ✅ DirectML | Recommended for Windows AI |
| PyTorch | ⚠️ | DirectML backend |
| Windows Studio Effects | ✅ | System-level AI features |
| Small LLMs (Phi/Granite) | ✅ | CPU+GPU hybrid inference |
Related Products
If you're evaluating alternatives, the following products may also fit your scenario:
- Intel NPU (Neural Processing Unit) — Intel (NPU Neural Processor)
- Apple Neural Engine — Apple (NPU Neural Processor)
- Qualcomm Hexagon NPU — Qualcomm (NPU Neural Processor)
- Huawei Ascend — Huawei (NPU Neural Processor)
- MediaTek NeuroPilot — MediaTek (NPU Neural Processor)
- NVIDIA GPU / CUDA — NVIDIA (GPU Graphics Processor)
- Groq LPU — Groq (LPU Language Processor)