Chip and graphic processing hardware manufacturer Nvidia used its GPU Technology Conference to introduce the newest member of the Jetson family to its customers.
Jetson Nano is described as an AI computer delivering 472 GFLOPS of compute performance as well as support for high-resolution sensors and frameworks such as TensorRT for deep learning.
The project is meant to make “AI more accessible to everyone” and “bring AI to the maker movement” to get people innovating. Other mini-computers seldomly come with the computing power needed to perform AI tasks in real-time, so it will be interesting to see what the Nano really will really bring to the table. Intel’s try at upping the ante on computing boards, the Intel Galileo, survived a bare four years and two versions before the company announced the end of life of the range.
Nvidia’s new pocket computer will begin shipping in June, cost $129 (in quantities of 1,000 or more). It will measure 6.96cm x 4.5cm and come with a quad-core ARM Cortex-A57 MPCore processor. It is supposed to sport a GPU in Nvidia Maxwell architecture with 128 CUDA cores, 16 GB eMMC 5.1 flash storage, 4 GB 64-bit LPDDR4 memory, three USB 2.0 as well as one USB 3.0 port. The Nano can be fitted with either a HDMI 2.0 connector or ports for DP1.2, eDP 1.4 or DSI and include 12 MIPI CSI-2 DPHY lanes.
Developers can get a cheaper developer kit ($99) with four USB 3.0 ports, USB 2.0 Micro-B, Gigabit Ethernet, M.2 Key E, ports for HDMI 2.0 and eDP 1.4, and a MIPI CSI-2 DPHY lane. For storage you’d have to get an additional microSD card and there isn’t any WiFi capability integrated.
Other than that, Nvidia provides an Nvidia AI platform compatible SDK called JetPack which is built on CUDA X and includes the latest versions of CUDA, cuDNN, TensorRT and a Linux OS.
To get “developers, makers, and enthusiasts” started, the company will also offer tutorials, a “collection of ready-to-use projects” and sources to build a small robot as a reference platform.