Advanced Computing in the Age of AI | Wednesday, April 24, 2024

Nvidia’s Jetson Flying to Shelves, Edge Servers 

Nvidia, well on its way to dominating AI in data centers with its chips, is trying to make inroads into devices on the edge.

The company in July will start shipping modules of its Jetson development boards, which it claimed deliver server-class AI performance in a computer that can be held in the palm of a hand.

The Jetson AGX Orin developer boards based on Nvidia's previous-generation Ampere-based GPUs, and is designed for use in robots, autonomous cars, medical devices and more. Orin is the central computing unit behind the company's Drive platform for autonomous cars, and will appear in vehicles from the likes of Volvo and Mercedes-Benz.

Jetson initially started as a developer board for hobbyists to make drones and robots. But as Nvidia builds out its AI data center infrastructure, Jetson is turning into an important platform for on-device and autonomous AI. Nvidia also sees Jetson boards feeding AI-related data from edge devices like automobiles and machines into the company's larger AI factories to train models and create datasets.

Nvidia has reached a milestone of over 1 million developers and over 6,000 companies developing and deploying products, and its partners are also involved in AI software, hardware, design services, sensors and security management. The developer boards are tuned for Nvidia’s CUDA proprietary software stack on which customers can build supercomputing and AI applications.

The Jetson AGX Orin with 32GB of RAM and eight CPUs based on Arm's Cortex-A78 design will ship in July for $899. The board delivers AI performance of 200 trillion operations per second (TOPS). The GPU has maximum frequency of 939MHz, and CPU tops out at 1.4GHz.

An upgraded Orin model with 64GB of RAM and 12 Cortex-A78 CPUs will ship in October this year for $1,599. The board has 2,048 CUDA cores and 64 Tensor Cores for AI applications. The GPU has a maximum frequency of 1.3GHz and CPU tops out at 1.6GHz.

The board delivers 275 TOPS of performance, which is about five times the performance over the previous generation product, said Amit Goel, director of product management for edge AI and robotics, in a press briefing held in conjunction the Computex trade show, taking place this week. The Jetson AGX Orin succeeds the Jetson board called Xavier.

The boards have PCIe Gen4 lanes, Gigabit Ethernet, DisplayPort and USB 3.2 interfaces. Each supports up to six camera interfaces for AI applications such as visual recognition and inspection.

Nvidia has filled out the lower-end lineup of the boards with a new developer module called Orin NX. The Orin NX with 16GB of memory and eight Cortex-A78 CPUs is priced at $599, and will be available in September this year. The Orin NX with 8GB of RAM and six Cortex-A78 CPUs will ship in December for $399.

Nvidia also announced the availability of the Jetson AGX Orin Developer Kit, introduced at the company’s GPU Technology Conference earlier this year. The developer kit includes the Orin model with 64GB of RAM and 12 Cortex-A78 CPUs.

"Thanks to a unified software architecture developers can use the Jetson AGX Orin developer kit to emulate any of the four products in the Jetson Orin portfolio," Goel said, referring also to the Orin NX boards.

The developer kit is priced at $1,999 on Arrow Electronics' website.

The company also announced that 30 partners worldwide would be bringing Jetson-based servers and appliances to market. The systems will come with different configurations, and multiple connectivity and interface options. The designs are used for commercial or ruggedized environments and intended for deployment in robotics, manufacturing, retail, transportation, smart cities and healthcare sectors, Goel said.

Many of those companies developing Jetson-powered rigs, including include Aaeon and Advantech, are based in Taiwan. Some partners will be showing Jetson-based products at this week's Computex trade show in Taipei.

Header caption: Jetson AGX Orin developer kit with production modules

EnterpriseAI