NPU & AI PC Hub
The brains of the AI PC revolution, on-device intelligence
Peak NPU Performance across 32 processors
What is an NPU?
Neural Processing Units are dedicated AI accelerators built directly into processors, handling trillions of operations per second while sipping power.
CPU vs GPU vs NPU
General-purpose cores. Flexible but inefficient for AI.
Massively parallel. Great for training, overkill for inference.
purpose-built for neural networks. Maximum efficiency.
Why NPUs Matter for AI
- 10-50x more efficient than CPU for AI workloads
- Run AI locally, no cloud, no privacy concerns
- Extended battery life during AI tasks
- Unlock Windows Copilot+ features
- Enable real-time image generation & video effects
Local LLMs
Run Mistral, Llama, Phi-3 privately on your device
Copilot+
Windows AI features powered by on-device NPU
Image Generation
Stable Diffusion & Cocreator in real-time
Video Enhancement
AI upscaling, background blur, eye contact
NPU Comparison
Compare all 32 NPU processors across Intel, AMD, Qualcomm, and Apple. Sort by TOPS, brand, or architecture.
| Brand | NPU Name | TOPS | Architecture | TDP | Process | Frameworks | Copilot+ | Buy |
|---|---|---|---|---|---|---|---|---|
 AMD | Ryzen AI 9 HX 370 Ryzen AI 300 (Strix Point) TSMC 4nm · Jul 2024 | 50 | Strix Point (XDNA 2) | 28W | TSMC 4nm | Ryzen AI SDKONNX RuntimeDirectML | Certified | Amazon |
| AMD | Ryzen AI 9 365 Ryzen AI 300 (Strix Point) TSMC 4nm · Jul 2024 | 50 | Strix Point (XDNA 2) | 28W | TSMC 4nm | Ryzen AI SDKONNX RuntimeDirectML | Certified | Amazon |
| AMD | Ryzen AI Max+ 395 Ryzen AI Max (Fire Range) TSMC 4nm · Jan 2025 | 50 | Fire Range (XDNA 2 + RDNA 3.5) | 55W | TSMC 4nm | Ryzen AI SDKONNX RuntimeDirectML | N/A | Amazon |
| AMD | Ryzen AI Max 390 Ryzen AI Max (Fire Range) TSMC 4nm · Jan 2025 | 50 | Fire Range (XDNA 2 + RDNA 3.5) | 55W | TSMC 4nm | Ryzen AI SDKONNX RuntimeDirectML | N/A | Amazon |
 Intel | Core Ultra 9 288V Core Ultra 200V (Lunar Lake) TSMC N3B · Sep 2024 | 48 | Lunar Lake | 30W | TSMC N3B | OpenVINOONNX RuntimeDirectML | Certified | Amazon |
| Intel | Core Ultra 7 268V Core Ultra 200V (Lunar Lake) TSMC N3B · Sep 2024 | 48 | Lunar Lake | 17W | TSMC N3B | OpenVINOONNX RuntimeDirectML | Certified | Amazon |
| Intel | Core Ultra 7 258V Core Ultra 200V (Lunar Lake) TSMC N3B · Sep 2024 | 47 | Lunar Lake | 17W | TSMC N3B | OpenVINOONNX RuntimeDirectML | Certified | Amazon |
 Qualcomm | Snapdragon X Elite X1E-84-100 Snapdragon X Elite TSMC 4nm · Jun 2024 | 45 | Oryon (Hexagon NPU) | 40W | TSMC 4nm | QNNONNX RuntimeDirectML | Certified | Amazon |
| Qualcomm | Snapdragon X Elite X1E-80-100 Snapdragon X Elite TSMC 4nm · Jun 2024 | 45 | Oryon (Hexagon NPU) | 40W | TSMC 4nm | QNNONNX RuntimeDirectML | Certified | Amazon |
| Qualcomm | Snapdragon X Plus X1P-66-100 Snapdragon X Plus TSMC 4nm · Jun 2024 | 45 | Oryon (Hexagon NPU) | 30W | TSMC 4nm | QNNONNX RuntimeDirectML | Certified | Amazon |
| Intel | Core Ultra 5 238V Core Ultra 200V (Lunar Lake) TSMC N3B · Sep 2024 | 40 | Lunar Lake | 17W | TSMC N3B | OpenVINOONNX RuntimeDirectML | Certified | Amazon |
| AMD | Ryzen AI Max 385 Ryzen AI Max (Fire Range) TSMC 4nm · Jan 2025 | 40 | Fire Range (XDNA 2 + RDNA 3.5) | 55W | TSMC 4nm | Ryzen AI SDKONNX RuntimeDirectML | N/A | Amazon |
| Qualcomm | Snapdragon X Plus X1P-46-100 Snapdragon X Plus TSMC 4nm · Jun 2024 | 40 | Oryon (Hexagon NPU) | 30W | TSMC 4nm | QNNONNX RuntimeDirectML | Certified | Amazon |
| AMD | Ryzen 7 8700G Ryzen 8000G (Hawk Point) TSMC 4nm · Jan 2024 | 39 | Hawk Point (XDNA) | 65W | TSMC 4nm | Ryzen AI SDKONNX RuntimeDirectML | N/A | Amazon |
| AMD | Ryzen AI 7 PRO 360 Ryzen AI 300 (Strix Point) TSMC 4nm · Jul 2024 | 38 | Strix Point (XDNA 2) | 28W | TSMC 4nm | Ryzen AI SDKONNX RuntimeDirectML | Certified | Amazon |
| AMD | Ryzen 5 8600G Ryzen 8000G (Hawk Point) TSMC 4nm · Jan 2024 | 38 | Hawk Point (XDNA) | 65W | TSMC 4nm | Ryzen AI SDKONNX RuntimeDirectML | N/A | Amazon |
 Apple | M4 Max Apple M4 TSMC 3nm · Oct 2024 | 38 | M4 (Neural Engine) | 50W | TSMC 3nm | Core MLMLXONNX Runtime | N/A | Amazon |
| Apple | M4 Apple M4 TSMC 3nm · Oct 2024 | 38 | M4 (Neural Engine) | 22W | TSMC 3nm | Core MLMLXONNX Runtime | N/A | Amazon |
| Apple | M4 Pro Apple M4 TSMC 3nm · Oct 2024 | 35 | M4 (Neural Engine) | 35W | TSMC 3nm | Core MLMLXONNX Runtime | N/A | Amazon |
| Apple | M3 Max Apple M3 TSMC 3nm · Oct 2023 | 35 | M3 (Neural Engine) | 50W | TSMC 3nm | Core MLMLXONNX Runtime | N/A | Amazon |
| Apple | M3 Pro Apple M3 TSMC 3nm · Oct 2023 | 35 | M3 (Neural Engine) | 35W | TSMC 3nm | Core MLMLXONNX Runtime | N/A | Amazon |
| Apple | M2 Ultra Apple M2 TSMC 5nm · Jun 2023 | 31.6 | M2 (Neural Engine) | 60W | TSMC 5nm | Core MLMLXONNX Runtime | N/A | Amazon |
| Apple | M3 Apple M3 TSMC 3nm · Oct 2023 | 18 | M3 (Neural Engine) | 22W | TSMC 3nm | Core MLMLXONNX Runtime | N/A | Amazon |
| Apple | M2 Max Apple M2 TSMC 5nm · Jan 2023 | 15.8 | M2 (Neural Engine) | 35W | TSMC 5nm | Core MLMLXONNX Runtime | N/A | Amazon |
| Intel | Core Ultra 9 285K Core Ultra 200 (Arrow Lake) TSMC N3B · Oct 2024 | 13 | Arrow Lake | 125W | TSMC N3B | OpenVINOONNX RuntimeDirectML | N/A | Amazon |
| Intel | Core Ultra 7 265K Core Ultra 200 (Arrow Lake) TSMC N3B · Oct 2024 | 13 | Arrow Lake | 125W | TSMC N3B | OpenVINOONNX RuntimeDirectML | N/A | Amazon |
| Intel | Core Ultra 5 245K Core Ultra 200 (Arrow Lake) TSMC N3B · Oct 2024 | 13 | Arrow Lake | 125W | TSMC N3B | OpenVINOONNX RuntimeDirectML | N/A | Amazon |
| Intel | Core Ultra 9 185H Core Ultra (Meteor Lake) Intel 4 (7nm) · Dec 2023 | 11 | Meteor Lake | 45W | Intel 4 (7nm) | OpenVINOONNX RuntimeDirectML | N/A | Amazon |
| Intel | Core Ultra 7 165H Core Ultra (Meteor Lake) Intel 4 (7nm) · Dec 2023 | 11 | Meteor Lake | 28W | Intel 4 (7nm) | OpenVINOONNX RuntimeDirectML | N/A | Amazon |
| Intel | Core Ultra 7 155H Core Ultra (Meteor Lake) Intel 4 (7nm) · Dec 2023 | 11 | Meteor Lake | 28W | Intel 4 (7nm) | OpenVINOONNX RuntimeDirectML | N/A | Amazon |
| Intel | Core Ultra 5 135H Core Ultra (Meteor Lake) Intel 4 (7nm) · Dec 2023 | 9 | Meteor Lake | 28W | Intel 4 (7nm) | OpenVINOONNX RuntimeDirectML | N/A | Amazon |
| Intel | Core Ultra 5 125H Core Ultra (Meteor Lake) Intel 4 (7nm) · Dec 2023 | 9 | Meteor Lake | 28W | Intel 4 (7nm) | OpenVINOONNX RuntimeDirectML | N/A | Amazon |
Ryzen AI 9 HX 370
Strix Point (XDNA 2) — 28W
Ryzen AI 300 (Strix Point)
Ryzen AI 9 365
Strix Point (XDNA 2) — 28W
Ryzen AI 300 (Strix Point)
Ryzen AI Max+ 395
Fire Range (XDNA 2 + RDNA 3.5) — 55W
Ryzen AI Max (Fire Range)
Ryzen AI Max 390
Fire Range (XDNA 2 + RDNA 3.5) — 55W
Ryzen AI Max (Fire Range)
Core Ultra 9 288V
Lunar Lake — 30W
Core Ultra 200V (Lunar Lake)
Core Ultra 7 268V
Lunar Lake — 17W
Core Ultra 200V (Lunar Lake)
Core Ultra 7 258V
Lunar Lake — 17W
Core Ultra 200V (Lunar Lake)
Snapdragon X Elite X1E-84-100
Oryon (Hexagon NPU) — 40W
Snapdragon X Elite
Snapdragon X Elite X1E-80-100
Oryon (Hexagon NPU) — 40W
Snapdragon X Elite
Snapdragon X Plus X1P-66-100
Oryon (Hexagon NPU) — 30W
Snapdragon X Plus
Core Ultra 5 238V
Lunar Lake — 17W
Core Ultra 200V (Lunar Lake)
Ryzen AI Max 385
Fire Range (XDNA 2 + RDNA 3.5) — 55W
Ryzen AI Max (Fire Range)
Snapdragon X Plus X1P-46-100
Oryon (Hexagon NPU) — 30W
Snapdragon X Plus
Ryzen 7 8700G
Hawk Point (XDNA) — 65W
Ryzen 8000G (Hawk Point)
Ryzen AI 7 PRO 360
Strix Point (XDNA 2) — 28W
Ryzen AI 300 (Strix Point)
Ryzen 5 8600G
Hawk Point (XDNA) — 65W
Ryzen 8000G (Hawk Point)
Core Ultra 9 285K
Arrow Lake — 125W
Core Ultra 200 (Arrow Lake)
Core Ultra 7 265K
Arrow Lake — 125W
Core Ultra 200 (Arrow Lake)
Core Ultra 5 245K
Arrow Lake — 125W
Core Ultra 200 (Arrow Lake)
Core Ultra 9 185H
Meteor Lake — 45W
Core Ultra (Meteor Lake)
Core Ultra 7 165H
Meteor Lake — 28W
Core Ultra (Meteor Lake)
Core Ultra 7 155H
Meteor Lake — 28W
Core Ultra (Meteor Lake)
Core Ultra 5 135H
Meteor Lake — 28W
Core Ultra (Meteor Lake)
Core Ultra 5 125H
Meteor Lake — 28W
Core Ultra (Meteor Lake)
AI PC Showcase
52 Copilot+ PCs and AI laptops in our database, ranked by on-device AI capability.
Surface Laptop 7 15"
Snapdragon X Elite
16 GB
512 GB
15" 2496x1664 120Hz Touch
Surface Laptop 7 13.8"
Snapdragon X Plus
16 GB
256 GB
13.8" 2304x1536 120Hz Touch
Surface Pro 11
Snapdragon X Plus
16 GB
512 GB
13" 2880x1920 120Hz Touch
ASUS Zenbook S 16
Ryzen AI 9 HX 370
32 GB
1024 GB
16" 2880x1800 120Hz OLED
ASUS Vivobook S 15
Snapdragon X Elite
16 GB
1024 GB
15.6" 2880x1620 120Hz OLED
ASUS ProArt P16
Ryzen AI 9
32 GB
1024 GB
16" 3840x2400 120Hz OLED Touch
Benchmark Results
Real-world NPU benchmark data from our AI PC test suite.
Local LLM, Phi-3 (3.8B)
ThroughputMicrosoft Copilot+
The next generation of Windows PCs. 40+ TOPS NPU required for Copilot+ certification.
Copilot+ Requirements
To qualify as a Copilot+ PC, a device must have an NPU capable of at least 40 TOPS, along with 16GB+ RAM and 256GB+ SSD storage. All devices in our comparison meet or exceed this threshold.
Recall
Search and retrieve anything you have seen on your PC using natural language.
Cocreator
Generate and refine images in real-time with AI-powered drawing assistance.
Live Captions
Real-time translation of 44+ languages into English captions for any audio.
Windows Studio Effects
AI-enhanced camera features: eye contact, background blur, and voice focus.
Software Compatibility
AI framework support across NPU platforms. Native support delivers the best performance.
| Framework | Vendor | Version | Intel | AMD | Qualcomm | Apple |
|---|---|---|---|---|---|---|
| ONNX Runtime | Microsoft | 1.18+ | Optimized | Optimized | Optimized | Supported |
| DirectML | Microsoft | 1.15+ | Optimized | Optimized | Optimized | N/A |
| OpenVINO | Intel | 2024.2+ | Native | Supported | Supported | N/A |
| Qualcomm AI Stack (QNN) | Qualcomm | 2.24+ | N/A | N/A | Native | N/A |
| Ryzen AI SDK | AMD | 1.3+ | N/A | Native | N/A | N/A |
| Core ML | Apple | 7.0+ | N/A | N/A | N/A | Native |
| MLX | Apple | 0.16+ | N/A | N/A | N/A | Native |
Which NPU is Right for You?
A decision guide based on your primary use case. Choose the path that matches your workflow.
Best for running large local LLMs, image generation, and video enhancement. The XDNA 2 architecture leads in raw throughput.
Still undecided? Compare all NPUs in detail with our interactive tools.