EVERYTHING YOU NEED TO KNOW ABOUT JETSON ORIN NANO-Part 1

May 17, 2023

Posted: May 17, 2023

Categories: Tech blogs

Tags: Jetson Orin Nano, Jetson Nano Orin

Everything You Need to Know about Jetson Orin Nano

Prologue

Nvidia’s range of Jetson boards will never be the typical Raspberry Pi alternative for makers. Aimed at entry-level AI-based robots, drones and cameras, the latest board, the $499 Jetson Orin Nano, boosts processing power while keeping the kit compact.

The Jetson Orin Nano 8GB module adds 1024 ampere-based CUDA cores to the 128 CUDA cores of the Nvidia Maxwell GPU. The additional cores and updated architecture mean that the Orin Nano delivers 80 times the AI performance of the Jetson Nano. The Jeston Orin Nano series starts with six Arm A78AE CPU cores, delivering nearly seven times the performance of the Jetson Nano. The Orin Nano uses the same AI architecture as the Jetson AGX Orin module but at a more affordable price.

What is Jetson Orin Nano?

Jetson Nano Orin is a strong product line in Nvidia’s Jetson family of small, powerful computers designed to power entry-level edge AI applications and devices.

We have seen the most powerful SoC on an embedded system in the world, offering unparalleled performance for the price, thanks to Jetson Orin Nano’s 6-core Arm Cortex-A78AE CPU and a serious Ampere-based GPU setup that offers 1028 CUDA cores and 32 third-generation Tensor cores, making it computationally the most powerful SoC we’ve seen.

If you noticed that the “AE” suffix was up there, it refers to a special core architecture that can accomplish SIL 3 safety for essential industrial applications with unique smodes. It is interesting to note that Tegra’s Orin series chips are reportedly the first to license this core — the Orin Nano is truly at the cutting edge of technology.

On board is 8GB of LPDDR5 RAM, shared by the CPU and GPU, offering an average memory bandwidth of 68GB/s. Although it isn’t as fast as GDDR5 or HBM2 RAM used in discrete GPUs, it shouldn’t cause any bottlenecking.

Hardware

The Jetson Orin Nano SOM is available in 4GB and 8GB versions, the latter offering twice the memory and GPU performance. The Developer Kit only includes the 8GB version. The compact carrier board features DisplayPort, USB, Gigabit Ethernet and a USB Type-C port. It also has two MIPI CSI connectors, a 40-pin GPIO header, a PWM header, and an external button control header. The board has expansion options, including a microSD slot, an M.2 Key E slot for Wi-Fi, and two M.2 Key M slots for PCIe. The CPU is a 6-core Arm Cortex-A78AE and the GPU is an NVIDIA Ampere with 1,024 CUDA cores and 32 Tensor cores. The board has no accelerators and supports software-only H.264 video encoding up to 3x 1080p30 and decoding up to 1x 4k60, 2x 4k30, 5x 1080p60 or 11x 1080p30. The board measures approximately 100x79x21mm.

Software

The Jetson Orin Nano runs JetPack, NVIDIA’s embedded software stack based on Ubuntu Linux. It was tested on a pre-release version of JetPack 5.1.1, which has minor bugs that should be fixed soon. JetPack 5.1.1 is based on Ubuntu 20.04.5 LTS and comes with various software, but it doesn’t include a way to train networks on the device. Therefore, training must be done off-device, which limits its potential as an affordable all-in-one workstation for AI development. However, the Jetson Orin Nano is supported by an impressive software stack, including the TAO Toolkit for training and the Omniverse Replicator for synthetic dataset generation. It supports two M.2 Key M PCIe slots for high-capacity storage and a pre-configured M.2 Key E slot for Wi-Fi connectivity.