Thought Leadership
NVIDIA is expanding its Jetson family with a particularly cost-effective model for AI applications. The new Jetson Orin Nano Super, priced at 249 US dollars, is designed to significantly lower the entry barrier for developers and smaller companies.
The small and compact computing unit is aimed specifically at developers of AI, robotics and computer vision. According to the manufacturer, the system, which is available immediately, achieves a 70 percent higher computing power of 67 INT8 TOPS. The memory bandwidth has been increased to 102 GB/s, which corresponds to an increase of 50 percent. In terms of AI inference performance, NVIDIA promises an improvement by a factor of 1.7 compared to the previous model. AI inference performance describes how quickly and efficiently a system can execute already trained AI models in practice. In other words, it is the ability to draw conclusions from a trained model and deliver results.
At the presentation of the new device, NVIDIA founder Jensen Huang surprised everyone with an unusual appearance: He presented the compact device on a tray – almost like a freshly baked product from the oven.
“We are reaching a tipping point where generative AI is finally arriving at the edge,” explained Deepu Talla, NVIDIA’s Vice President of Robotics and Edge Computing. “Edge” refers to decentralized devices in factories and laboratories, as opposed to the large data centers where AI computations have mainly taken place so far.
Although the Jetson Orin Nano Super is equipped with less advanced chips than NVIDIA’s high-end products, it should still be powerful enough to run current generative AI models. The more expensive Jetson Thor model is still intended for more demanding applications such as humanoid robots and complex automation systems.
Nvidia focuses mainly on large companies and AI start-ups that invest billions in hardware for training and operating their AI models. With the Jetson product series, however, the company is now taking a different approach: these affordable devices are aimed specifically at smaller companies, hobby developers and students who want to develop AI-based products themselves.
