ProductsGateworksGW16168 AI Accelerator

Gateworks

GW16168 AI Accelerator

USA-made M.2 card with the NXP Ara240 neural processor, adding up to 40 eTOPS of AI to VeniceFLEX and Catalina boards.

  • NXP Ara240 DNPU
  • Up to 40 eTOPS
  • 8 or 16 GB LPDDR4
  • M.2 2280, 6.2 g
GW16168 AI Accelerator

Overview

The GW16168 is an M.2 M-Key 2280 accelerator card built around the NXP Ara240 discrete neural processing unit. It adds up to 40 eTOPS of AI inference to a Gateworks single board computer, so vision models and language models run on the device itself, with no cloud connection. The card carries 8 or 16 GB of LPDDR4, weighs 6.2 g and draws about 6.6 W. Gateworks designs and manufactures it in the USA, with the USA as its country of origin.

Built for

  • Onboard vision for drones and robots
  • Object detection and video analytics
  • Language and vision-language models at the edge
  • Rugged AI systems with no cloud link

Key features

40 eTOPS

The NXP Ara240 DNPU delivers up to 40 equivalent tera operations per second.

Large models on the card

8 or 16 GB of LPDDR4 holds vision models and language models of up to 16 billion parameters, or larger with INT4 quantization.

Common frameworks

Models come from TensorFlow, PyTorch or ONNX through the SDK compiler for CNNs and LLMs.

Small and light

The M.2 2280 card measures 22 x 80 x 3.5 mm and weighs 6.2 g.

Secure boot

Secure boot and a root of trust are built in.

Made in the USA

Designed and manufactured by Gateworks in California, with the USA as its country of origin.

Specifications

Processor
NXP Ara240 DNPU
AI performance
Up to 40 eTOPS
Memory
8 or 16 GB LPDDR4
Interface
PCIe Gen4 x4, works with Gen3 hosts
Form factor
M.2 M-Key 2280, 22 x 80 x 3.5 mm
Weight
6.2 g
Power
3.3 V from the M.2 connector, 2.0 A typical
Frameworks
TensorFlow, PyTorch, ONNX
Host boards
VeniceFLEX GW8100, GW8200, Catalina GW9200, GW9400
Development kit
GW11062-1 with a Catalina GW9200
Operating temp
-40 to +85 °C (-25 to +85 °C on GW16168-SP521)
Country of origin
USA

Models

ModelMemoryOperating temp
GW1616816 GB LPDDR4-40 to +85 °C
GW16168-SP5208 GB LPDDR4-40 to +85 °C
GW16168-SP52116 GB LPDDR4-25 to +85 °C

Specifications as published by Gateworks. Configurations vary by model, so we will confirm the exact build with your quote.

Buying the GW16168 AI Accelerator in India

Authorised Gateworks distributor

Genuine Gateworks products, sourced directly from Gateworks.

GST invoice

Invoiced in India with GST, so registered businesses can claim input tax credit.

From stock or to order

Some products ship from our stock in India, and we order the rest from Gateworks for you.

Help choosing

Tell us the application and we will recommend the right Gateworks product.

Need help choosing?

Tell us your application. We will recommend the right Gateworks product, quote it and support it after delivery.