The new nano version of the Jetson Embedded Developer Kit is now available for $ 99. The production version will be available from June in the trade and cost $ 129 at the purchase of thousands of units. She comes on a 45 × 70 millimeter small board. The complete developer kit measures 80 × 110 mm. It should make do with a maximum of 10 watts and also offer an option for 5-watt operation. Power is supplied via 5 V DC – also via micro-USB.
Especially fast only with AI
It includes a quad-core processor with ARM-A57 cores and 4 GB of LPDDR4 RAM with a transfer rate of 25.6 GB / s. At the heart of Nvidia’s vision is the integrated GPU with 128 shader cores and the older Maxwell architecture. The should be with 472 GFLOPS (billion computing steps per second) particularly suitable for AI tasks. The specified throughput is achieved only with FP16 accuracy, which is sufficient for some AI calculations.
There are four USB-3 ports, a CSI-2 camera port, HDMi 2.0 and DP1.3, and Gigabit Ethernet. Data can be stored on an SSD in the M.2 slot. The video decoders handle the data stream of a 4K video at 60 frames per second – alternatively two at 30 fps or four times in 1080p.
In addition to the omnipresent keyword KI, the platform is also aimed at hobbyists, who also have access to a 40-pin GPIO.
Jetson SDK 4.2 with Ubuntu 18.04
The Jet-Pack 4.2 SDK (currently version 4.1.1) contains a complete Linux environment (Ubuntu 18.04) including Nvidia’s programming environment CUDA 10 and AI libraries cuDNN 7.3 and Tensor RT. TensorFlow, PyTorch, Caffe, and others should also be easy to install and use. For example, the operating system can be accommodated on a microSD card from which Jetson Nano also boots thanks to the integrated reader directly.