Gateworks and NXP Semiconductors have unveiled the GW16168 AI acceleration card at Embedded World, introducing a modular solution designed to simplify the deployment of artificial intelligence at the edge. The new hardware is intended to help engineers and businesses integrate high-performance AI capabilities without redesigning their entire computing platforms.
Deploying in-house AI systems can require substantial investments in infrastructure, engineering resources and system redesign. In many cases, introducing new AI hardware forces organisations to rethink their entire hardware stack—from single board computers to thermal management solutions—creating both cost and complexity. The GW16168 has been developed to address these challenges by enabling a modular and flexible approach to AI acceleration.
The card is based on what Gateworks describes as a “decoupled AI architecture.” This design philosophy allows developers to add AI capabilities to existing hardware platforms without replacing otherwise capable computing systems. By separating the AI processing element from the primary compute platform, organisations can upgrade AI capabilities independently of their existing hardware.
The GW16168 is built in an M.2 2280 M-Key form factor and incorporates NXP’s Ara240 deep neural processing unit (DNPU). The module includes 16 GB of LPDDR4 memory, allowing AI inference workloads to run directly on the card rather than overloading the host processor. Offloading these workloads enables the host system to focus on application logic, data management and I/O operations.
Through this architecture, the module can be added to platforms such as those based on NXP’s i.MX 8M Plus or i.MX 95 application processors using the M.2 interface. Systems that would normally operate at full capacity when running AI inference can now distribute processing tasks more efficiently.
The software ecosystem supporting the module includes the Ara240 SDK, which provides a complete development environment including compiler tools and compatibility with common AI frameworks such as TensorFlow, PyTorch and ONNX. Model conversion tools and optimisation utilities allow developers to adapt existing AI models for deployment on edge devices.
Ravi Annavajjhala, Vice President and General Manager of Neural Processing Units at NXP Semiconductors, noted that the GW16168 demonstrates how modular AI architectures can improve flexibility and scalability in edge computing. By combining NXP’s AI processing technology with Gateworks’ industrial design expertise, organisations can enhance AI performance without major system redesigns.
The GW16168 delivers up to 40 eTOPS of AI performance while operating within a compact power envelope. The module and its associated development kit will be available through distributors including DigiKey, Braemac, RoundSolutions and Avnet, with shipments expected to begin in late May.
