Axelera Europa AIPU
Vendor: Axelera AI (Netherlands)
Category: ASIC Dedicated Accelerator
Architecture: D-IMC (Digital In-Memory Compute)
Introduction
Axelera Europa is the second-generation AI processor unit (AIPU) from Dutch AI hardware company Axelera AI, based on digital in-memory computing (D-IMC) technology. Europa features 8 second-generation AI cores, each integrating advanced D-IMC technology and a large vector engine, delivering 629 TOPS INT8 compute, designed for edge-to-enterprise multimodal AI inference applications.
Specifications
| Model | Compute | Memory | Interface | TDP | Process |
|---|---|---|---|---|---|
| Europa AIPU | 629 TOPS (INT8) | 128MB on-chip SRAM + LPDDR5 | PCIe / M.2 | Low power | Advanced node |
| Metis M.2 Max | Edge inference optimized | M.2 form factor | M.2 | Ultra-low power | Advanced node |
Official Website
Driver Downloads
Linux
Related Documentation
OS Support
| Windows | Linux | macOS | Android |
|---|---|---|---|
| ❌ | ✅ | ❌ | ⚠️ (Embedded) |
Version History
| Version | Release Date | Description |
|---|---|---|
| Europa AIPU | 2026 H1 shipping | 2nd-gen AI core, 629 TOPS |
| Metis M.2 Max | 2025 | M.2 form factor edge inference |
Performance Benchmarks
| Model | Task | Performance Metric |
|---|---|---|
| Europa | INT8 Inference | 629 TOPS |
| Europa | Energy Efficiency | 3-5× higher than comparable products |
Pricing
| Model | Reference Price | Notes |
|---|---|---|
| Europa | Contact vendor | Enterprise/edge deployment |
| Europa PCIe | Contact vendor | Single-chip 16GB to quad-chip 256GB |
Quick Installation
Linux
# Install Voyager SDK
# See official documentation
Code Examples
Python (Axelera SDK)
import axelera as ax
# Load quantized model
model = ax.load_model("model.axn")
output = model.run(input_data)
Architecture Highlights
- D-IMC (Digital In-Memory Compute): Eliminates memory bottlenecks by performing computation directly in storage
- 8 AI Cores: Each core integrates D-IMC and a large vector engine
- 2 RISC-V Vector Clusters: 8 dedicated RISC-V vector processing cores, 4,915 GOPS peak
- Integrated H.264/H.265 Decoder: Hardware media decoding, offloading host CPU
Model Compatibility
| Model/Framework | Support | Notes |
|---|---|---|
| Quantized INT8 Models | ✅ Native | Best performance |
| LLM Inference | ✅ | Multi-user generative AI |
| Computer Vision | ✅ | Object detection/segmentation |
| Embedded AI | ✅ | Robotics/autonomous driving |
Related Products
If you're evaluating alternatives, the following products may also fit your scenario:
- Hailo-8 / Hailo-15 — Hailo (NPU Neural Processor)
- Google Edge TPU — Google (TPU Tensor Processor)
- FuriosaAI RNGD — FuriosaAI (ASIC Dedicated Accelerator)
- NVIDIA GPU / CUDA — NVIDIA (GPU Graphics Processor)
- d-Matrix Corsair — d-Matrix (ASIC Dedicated Accelerator)
- SambaNova RDU — SambaNova (ASIC Dedicated Accelerator)
- Cambricon Siyuan MLU — Cambricon (ASIC Dedicated Accelerator)