FAI-ARA240-M Edge AI Accelerator
The FAI-ARA240-M Edge AI Accelerator, a flagship solution from the Forlinx & NXP Gold Partnership, is a high-performance M.2 2280 module designed for real-time Generative AI. Powered by the dedicated NXP ARA-240 DNPU with 40 eTOPS and 16GB LPDDR4, it delivers an impressive 14 tokens/s for Llama2-7B via PCIe Gen4 x4, bringing server-grade intelligence to NXP-based industrial and embedded systems.
Why Choose FAI-ARA240-M?
- Flagship AI Performance: Powered by the ARA-240 DNPU, delivering 40 eTOPS optimized specifically for Transformers, LLMs, and VLMs.
- Industrial-Grade Efficiency: Ultra-low 12W power consumption—achieving the same compute power at 1/3 the energy cost of traditional solutions.
- Universal Plug-and-Play: Standard M.2 2280 (M-Key) with PCIe Gen4 x4, enabling instant AI upgrades for x86 and ARM platforms.
- High-Capacity Memory: Up to 16GB LPDDR4 and high-bandwidth architecture to support local deployment of large-scale models with ultra-low latency.
- Seamless Deployment: Full compatibility with PyTorch, TensorFlow, and ONNX, supported by a comprehensive SDK for rapid cross-platform porting.
New NXP DNPU
The NXP Ara240 offers up to 40 eTOPS of AI computing power and supports models such as CNN, VLM, LLM,
and VLA. It has a maximum memory capacity of 16GB and can easily function as a co-processor with various mainstream host platforms.
Powerful Computing Capability
It features the NXP Ara240 high-performance DNPU, providing up to 40 eTOPS of AI computing power and 16GB of memory. It effectively supports complex AI inference tasks at the edge, enhancing efficiency in applications like industrial multimodal sensing and real-time visual processing.
Wide Range of Models
It is compatible with major AI architectures like CNN, Transformers, and GenAI, catering to both traditional vision algorithms and advanced generative AI applications. It efficiently loads complex models, making it ideal for diverse industrial scenarios and highly extensible.
Rich AI Software Development
It provides robust software development tools, featuring an extensible compiler that supports CNNs, Vision Transformers, and large language models for efficient deployment. It works with INT4, INT8, and MSFP16 data types, using flexible quantization and data flow optimization to balance computing power, storage, and accuracy.
M.2 Packaging
Designed with a standard M.2 form factor, it easily integrates with mainstream host platforms, allowing for efficient scaling of AI computing power without complex modifications, thus reducing development costs.
Excellent Heat Dissipation
Optimized thermal architecture ensures stable temperature management under high-load industrial conditions, effectively preventing overheating and performance loss. This guarantees reliable 24/7 operation even in demanding environments.
Continuous Updates of User Documentation
Wide Industry Applications
Delivering robust performance across
industrial control , power, new energy, transportation, and healthcare sectors,
this solution combines high computing power and broad model compatibility with dedicated after-sales support from Forlinx Embedded.
Together, these advantages shorten your product's time to market and help you maintain a competitive edge.
▊ Hardware Features
| Hardware Parameters of Ara240 AI Acceleration Card | |
|---|---|
| Processor | NXP Ara240 |
| RAM |
|
| Interface | M.2 M-Key |
| Dimensions | M.2 2280 (22mm x 80mm) |
| Supported Host SoC |
|
| AI Performance | 40 eTOPS (equivalent TOPS) |
| Operating System | Linux |
| AI-Model Frameworks Supported |
|
| AI Model Architectures Supported |
|
Order Inquiry
Fill out the form below for volume pricing or technical specifications. We'll get back to you within 24 hours.
▊ How To Buy & Shipment
1. Order Online: We have an online store on Alibaba and DigiKey. Please contact us for details.
2. Order Offline: Send your inquiry to [email protected].
3. Payment Terms: 100% T/T in advance.
1. Delivery: Goods shipped by express upon receipt of payment.
2. Lead Time: Sample orders within 5 working days; bulk orders 6 weeks.
3. Shipping Charge: Buyers bear the shipping cost.
▊ Forlinx Recommended Products
Architecture: 6× [email protected] + 1x Cortex-M7@800MHz + 1x Cortex-M33@333MHz
Frequency: 333MHz ，800MHz，1.8GHz
RAM: 8GB LPDDR4x
ROM: 64GB eMMC
Architecture: 6× [email protected] + 1x Cortex-M7@800MHz + 1x Cortex-M33@333MHz
Frequency: 333MHz ，800MHz，1.8GHz
RAM: 8GB LPDDR4x
ROM: 64GB eMMC
Architecture: 4x [email protected]+1x Cortex-M7@800MHz
Frequency: 1.6GHZ
RAM: 1GB/2GB/4GB LPDDR4
ROM: 8GB/16GB eMMC
