Engineered for real-time, high-throughput AI workloads in intelligent transportation and smart parking applications, the system now delivers high AI inference performance due to NVIDIA’s latest high-performance operating mode.
As demand for smarter, more responsive traffic and parking systems rises, system integrators are challenged to deploy Edge AI solutions able to simultaneously handle multi-stream video analytics, licence plate recognition, object detection, and predictive insights—all with strict latency constraints. With the Super Mode feature enabled on NVIDIA Jetson Orin Nano, Axiomtek’s AIE100-ONA and AIE110-ONA now provide a notable performance leap to address these real-world deployment needs.
Super Mode maximises CPU and GPU utilization on the Orin Nano platform, enabling more advanced AI model execution with greater parallelism and efficiency. This means each Edge device can now support more camera streams and heavier AI models—reducing infrastructure costs while enhancing detection accuracy and response speed.
In response to the growing demand for smarter city deployments, Axiomtek’s internal teams re-evaluated performance-per-dollar across compact Edge AI systems.
“Our clients are looking for more compact systems that can handle heavier AI workloads without raising system cost or complexity,” said Ken Pan, Product Manager, Axiomtek.
This triggered a strategic shift in development focus.
Robin Lin, Head of Intelligent Automation Software Department, added: “We asked ourselves: what delivers more value per watt, per dollar, and per deployment cycle?”
The result was a renewed focus on fully unlocking Super Mode as a key enabler of performance scaling.
To meet this challenge, Vincent Chen, Senior Engineer of the AI Software R&D Department, proposed leveraging Super Mode as a performance breakthrough to unlock more computing potential from the Jetson Orin platform.
“In our DeepStream SDK-based testing with optimized YOLO models via TensorRT, we observed a 1.5–2x increase in throughput with Super Mode enabled,” said Chen. “Workloads that previously required two devices can now run on a single AIE100-ONA, while maintaining real-time response. Alternatively, developers can swap in larger models (e.g., YOLOv8n → YOLOv8s) to gain further mAP improvements, without changing deployment topology. This translates to a more cost-effective and scalable strategy, helping customers maximise performance per device and reduce total system investment.”
Reflecting on the development journey, Lin concluded: “Super Mode is more than a performance boost—it’s a strategic enabler for our customers. It allows them to run larger models, consolidate devices, and simplify deployments, all while maintaining real-time responsiveness. It’s how we turn cutting-edge AI into practical, scalable solutions at the Edge.”
