NVIDIA recently debuted the Jetson Nano 2GB developer kit. For $59, the kit includes a credit-card-sized single-board computer with a quad-core ARM CPU and a 128 core Maxwell GPU. It comes with the JetPack SDK, a Ubuntu Linux-based developer SDK, and comprehensive documentation. The Jetson Nano 2GB kit also includes online training and certification, making it an ideal developer’s kit for students and new developers who just started AI programming.
Deep learning is one of the most notable advancements in computer science in the past decade. Deep learning-based AI applications are now used everywhere, from computer vision to speech to natural language processing. However, as developers, learning AI skills for professional programming has been difficult. The barrier is the GPU. Most deep learning algorithms and frameworks are designed to run on the GPU. They are too slow on the CPU. Yet, most computers are CPU-based. While a personal computer typically contains a GPU to drive graphics, the operating system and software stack in the computer are designed to "hide" the GPU and only use it to drive the display graphics.
When a developer writes a deep learning application and runs it on a personal or cloud-based computer, there is a good chance the program is actually running on CPUs. With the Jetson series of devices and software SDKs, NVIDIA creates a coherent development environment to learn and develop GPU-based AI applications. The JetPack SDK provides a customized version of eLinux, which is based on and compatible with Ubuntu 18.04, and curated versions of key software packages, such as Python, TensorFlow, PyTorch, Numpy, OpenCV, to ensure that the whole software stack is optimized for the GPU and ARM CPU hardware on the development board. In addition to official learning resources from NVIDIA, there is a vibrant community of developers and hobbyists in the Jetson ecosystem. There is a wealth of YouTube videos, open-source projects, and online articles for these devices.
The entry-level Jetson device, called the Jetson Nano, was priced at $99, which is a little high compared with other single board computers such as the Raspberry Pi, which is priced at $40 without the GPU. The new Jetson Nano 2GB's $59 price point is much more reasonable for price-sensitive students and hobbyists learning AI programming.
Like the regular Jetson Nano, the Jetson Nano 2GB has a 64-bit quad-core ARM A57 CPU clocked at 1.43 GHz, and a 128 CUDA core Maxwell GPU. The GPU delivers 472 GFLOPS computing power for AI applications. In fact, for AI applications, the Jetson Nano 2GB is 8 to 73 times faster than the most advanced Raspberry Pi 4.
The Jetson Nano 2GB board has several USB 2/3 connectors, a power connector, an HDMI display connector, an ethernet connector, GPIO pins, a camera kit connector, as well as an M2 key E connector for a WiFi and Bluetooth card. The 2GB refers to the on-board memory space. You do need an additional microSD card for the operating system and files in order to boot up and use the Jetson Nano 2GB. Furthermore, you will need to connect the card to an HDMI display, keyboard, and mouse, as well as the network (either cable-based Ethernet or WiFi card) before you can use it as a computer.
With its small size and low cost, the Jetson Nano 2GB can power computer vision applications in robots or drones. Its GPUs can analyze video streams from the camera in real-time, recognize objects and faces in each video frame, and send out corresponding control commands through its GPIO pins or USB connectors. However, with the 10W power consumption of the GPU working in full video image recognition mode, it is also challenging to keep the device running on batteries for an extended period of time. Therefore, the Jetson Nano 2GB is indeed primarily a learning device.
You can pre-order the Jetson Nano 2GB from online retailers, and it should be available in late October. Follow the learning resources and tutorials from the NVIDIA website to start programming AI applications on your GPU!