REDDIT · REDDIT// 31d agoINFRASTRUCTURE

Gateworks GW16168 brings 40 eTOPS edge AI

Gateworks and NXP have launched the GW16168, an M.2 AI acceleration card built around the Ara240 discrete NPU with up to 40 eTOPS, 16GB of onboard LPDDR4, and support for TensorFlow, PyTorch, and ONNX via the NXP Ara SDK. It is aimed at industrial and embedded systems that need to offload vision and LLM inference without redesigning the whole host platform.

// ANALYSIS

This is the kind of edge AI hardware launch that matters more to deployers than headline-chasing model news: it promises real inference throughput in a tiny, fanless-friendly module instead of another dev-board science project.

–The M.2 2280 form factor and PCIe Gen4 x4 interface make it a practical drop-in upgrade path for existing embedded and industrial systems
–Gateworks’ “decoupled AI architecture” pitch is credible because the card carries its own 16GB memory and offloads inference from the host CPU
–Support for PyTorch, TensorFlow, and ONNX through the Ara SDK lowers the barrier for teams already sitting on trained models
–The claimed 30B-parameter INT4 support is ambitious for edge deployments and positions the card well beyond basic vision-only accelerators
–Late-May availability through distributors is a good sign this is a real product launch, but missing pricing keeps the commercial story incomplete

// TAGS

gateworks-gw16168inferenceedge-aigpu

DISCOVERED

31d ago

2026-03-11

PUBLISHED

31d ago

2026-03-11

RELEVANCE

7/ 10

AUTHOR

DeliciousBelt9520