Autonomous machines, robotics and artificial intelligence have been transforming daily life, making it more convenient, safe and efficient. These technologies are being applied to multiple applications in domains such as industrial, automotive, healthcare, smart city, smart buildings, and entertainment among others. The ever-evolving technology landscape and the increasing need for real-time data processing and inference at the edge is encouraging SoC/GPU manufacturers to build powerful edge computing devices that can run modern Deep Learning (DL) workloads.

NVIDIA Jetson System on Modules (SoMs) are embedded AI computing platforms designed to accelerate edge AI product lifecycle. These high-performance, low-power compute engines are built to run complex computer vision, robotics, and other low-power applications. In addition, NVIDIA offers CUDA-X libraries, a set of highly optimized, GPU-accelerated libraries. CUDA-X offers a robust programming environment that works seamlessly with all Jetson SoMs, enabling customers to seamlessly use their application across various Jetson Platforms.

Advantages of NVIDIA Jetson Platform:

1) Versatile Hardware – The NVIDIA Jetson portfolio provides a wide range of high-performance, low-power system on modules, that feature multi-core ARM cores, NVIDIA GPU, AI computational performance up to 275 TOPS and a host of high-speed interfaces, ideal for complex DL pipelines.

2) Unified Software offers Ease of Development – NVIDIA Jetson System on modules and Development Kits are supported by a unified software architecture that helps developers to expand or enhance their product portfolio on different Jetson modules without further coding. This unified approach helps enterprises to create product portfolios with a significant return on investment.

3) Cloud Native Support for Scalable Deployments – Cloud-native technologies and workflows like orchestration and containerization are supported by all NVIDIA Jetson platforms, which accelerate edge AI product development, agile deployment, and management throughout the product lifecycle. Frequent updates and constant upkeeps are imperative to rapidly improve AI efficiency.

Here’s a sneak peek of the hardware capabilities of various Jetson Modules:

1) The NVIDIA Jetson Nano is a small, power-efficient, high-performance AI computing device to run AI applications, process data from high-resolution sensors and run multiple neural networks in parallel. This module is an ideal entry-level option to add advanced capabilities to embedded products and elementary solutions like intelligent gateways or entry-level NVRs. The Jetson Nano has an integrated 128-core Maxwell GPU, quad-core ARM A57 64-bit CPU, 4GB LPDDR4 memory, and support for MIPI CSI-2 and PCIe Gen2 high-speed I/O.

2) The NVIDIA Jetson TX2 is a power-efficient (7.5W) embedded AI computing device that offers up to 2.5x the performance of Jetson Nano. The module features a variety of standard hardware interfaces that makes it ideal for designing a wide range of small form-factor products that run advanced machine learning applications. The Jetson TX2 modules integrate NVIDIA Pascal™ GPU architecture with 256 NVIDIA CUDA cores, dual-Core NVIDIA Denver 2 64-Bit CPU, 8GB 128-bit LPDDR4 memory, and 32GB eMMC 5.1 storage. In addition, TX2 offers USB 3.0 Type A, USB 2.0 Micro USB (supports recovery and host mode), HDMI, M.2 Key E, PCI-E x4, Gigabit Ethernet, full-size SD, SATA Data and Power, GPIOs, I2C, I2S, SPI, CAN, TTL UART with flow control, display expansion header, and camera expansion header.

3) The NVIDIA Jetson Xavier NX is a small form-factor module, in the size of a credit card that offers up to 21 TOPs of accelerated AI computing. This module can process data from various high-resolution sensors and run multiple advanced neural networks in parallel, which is critical to modern embedded systems. The 384-core NVIDIA Volta™ GPU with 48 Tensor Cores offers high computing power for compact, intelligent machines like autonomous drones and portable medical devices. The module comes with 6-core NVIDIA Carmel ARMv8.2 CPU, dual DLA engines, 8 GB memory, 6 CSI camera and support a host of display, connectivity, and networking interfaces.

4) The NVIDIA Jetson AGX Xavier is a power-efficient module designed for autonomous machines that delivers up to 32 TOPS of AI performance. This compact module includes high-speed I/O and hardware acceleration for the entire AI pipeline enabling developers to come up with the latest AI applications. The Jetson AGX Xavier also provides an extended temperature range, vibration and shock specifications, making it one of the best choices for safety-certified and industrial-grade products. This SoM comes with NVIDIA Volta architecture GPU with 512 CUDA cores, 8-core NVIDIA Carmel Armv8.2 64-bit CPU and support up to 6 CSI cameras in addition to the host of the display, connectivity and networking interfaces.

5) With the launch of the new Jetson Orin Nano SoM, NVIDIA Jetson is setting a new standard for entry-level edge AI and robotics. Orin Nano comes in two variants, and it can deliver up to 80x the performance over the previous generation, up to 40 TOPS of AI performance, and comes with 5W and up to 15W power options. Jetson Orin Nano includes a video decode engine, 6-core Arm Cortex-A78AE CPU, ISP, audio processing engine, video image compositor, and video input block. This new SoM series also features up to 1024-core NVIDIA Ampere architecture GPU with 32 Tensor cores, supports FP16 and Int 8, and comes with 4GB and 8GB. This SoM comes with a strong Ecosystem, better software support, enhanced encoding capabilities, and memory bandwidth and supports external NVME for data storage directly as it has no eMMC.

6) The NVIDIA Jetson Orin NX provides high-speed interface support for multiple sensors, offers twice the CUDA cores, 5x performance of NVIDIA Jetson Xavier NX and 3X the performance of Jetson AGX Xavier for power-efficient autonomous machines. This module offers up to 100 TOPS of AI performance, 1024-core NVIDIA Ampere architecture GPU with 32 Tensor Cores, power configurable between 10W and 25W, supports external NVMe, shares the form factor compatibility with NVIDIA Jetson Xavier NX, and can give a tremendous performance in an amazingly compact package.

7) The Jetson AGX Orin is the latest incorporation from NVIDIA among all Jetson Modules. This SoM is one of the most powerful embedded processors currently in market, with up to 275 TOPs performance. The Jetson AGX Orin offers a giant leap forward in Edge AI and robotics. Jetson AGX Orin offers 8x performance over Jetson AGX Xavier with configurable power between 15W and 50W, making it ideal for autonomous machines and advanced robots. It will also help developers to deploy complex and large AI models to solve problems such as 3D perception, natural language understanding, and multi sensor fusion. This SoM supports multiple concurrent AI inference pipelines with onboard 64GB eMMC, 204 GB/s of memory bandwidth, and high-speed interface support for multiple sensors.

Table 1: NVIDIA Jetson Modules _ Spec Comparison

Conclusion

The NVIDIA Jetson System on Modules are power-performance-optimized edge devices that can transform every industry, including healthcare, retail, manufacturing, transportation, construction, logistics, agriculture, robotics, surveillance, and more. AI-based embedded product companies and developers are using NVIDIA Jetson SoMs to enhance and accelerate product development, demonstrating how cutting-edge technologies can facilitate new levels of success every day.

Mistral is a Member of the NVIDIA Partner Network, and our solutions facilitate accelerated product development, prototyping and reduce time to market. Our team has unparalleled expertise and design competency in offering custom Carrier Boards, Development Platforms and Application Boards based on Jetson System on Modules. Mistral also offers feature-rich, performance-oriented off-the-shelf Carrier Boards and AI-Sensor Fusion platforms compatible with the Jetson Nano, Jetson Xavier NX and Jetson TX2 NX SoMs. To know more, visit mistralsolutions.com/neuron.

Stay tuned for the second part of the blog that features various use cases of Jetson platforms!