Ara-2: Discrete NPU for Real-Time Generative AI

ARA-2-DNPU
Preproduction
  • This page contains information on a preproduction product. Specifications and information herein are subject to change without notice. For additional information please contact your sales representative.

Roll over image to zoom in

The Ara-2 discrete neural processing units (DNPUs) enable real-time Generative AI, Large Language Models (LLMs) and Vision Language Models (VLMs) execution on AI-enabled compute and embedded systems, delivering low latency, lower operational costs and enhanced data privacy. Its innovative architecture combines balanced compute, large on-chip memory and high off-chip bandwidth to efficiently execute large models.

Block Diagram

Choose a diagram:

NXP Applications Processor and Ara DNPU Connection

ARA-1-DNPU-BD-2

Integration of Ara DNPU and Applications Processors Host with Drivers

ARA-1-DNPU-BD-3

Machine learning deployment flow

ARA-1-DNPU-BD
Get diagram PDF

Features

AI model frameworks supported

  • TensorFlow
  • PyTorch
  • ONNX

AI Model Architectures Supported

  • Convolutional Neural Networks (CNNs)
  • Transformer Models
  • Large Language Models (LLMs)
  • Multimodal Language Models (MMLMs)
  • Vision Language Models (VLMs)
  • Vision Language Actions (VLAs)

Performance

  • Up to 40 eTOPS*

Security

  • Secure Boot
  • Root-of-trust processor

Memory Interface

  • Up to 16 GB LPDDR4(X)

Operating system support (Runtime)

  • Linux
  • Windows

Host interface

  • 4-lane PCIe Gen4
  • USB3 Gen 2

Chip package

  • 17 mm x 17 mm FCBGA

*eTOPS = equivalent TOPS

Documentation

Quick reference to our documentation types

Filters

1 documents

Compact List

Design Resources

Hardware

Quick reference to our board types.
Filters

2 hardware offerings

Software

Quick reference to our software types.

Filters

2 software files

Note: For better experience, software downloads are recommended on desktop.

Support

What do you need help with?

  • Narrow your search
Search