"We are convinced that we will help shape the future of industrial automation with the AI embedded computer," explains Florian Egger, Head of Sales at Syslogic. Accordingly, Syslogic is working at full speed to develop a complete deep learning solution together with a software partner. In addition, Syslogic hardware engineers are already planning an AI box PC with the new Nvidia Xavier platform - however, the company is relying on the Jetson TX2i module from Nvidia for its AI debut.

At the heart of the Jetson TX2 is the ARM SoC Tegra X2 with the code name Parker. It combines two computing cores with the Denver 2 microarchitecture developed by Nvidia itself with four Cortex-A57 cores and a Pascal GPU. The latter has 256 shader cores. The embedded computer therefore supports CUDA applications (CUDA: Compute Unified Device Architecture) and is suitable for various AI sub-areas such as machine vision, intelligent control and deep learning. Thanks to optional Wi-Fi, GPS and LTE functions, the AI computer can be easily integrated into the Industrial Internet of Things (IIoT).

Axiomtek also relies on Nvidia's Jetson-TX2 module with up to 8 GB LPDDR4 memory and an 802.11ac Wi-Fi module with Bluetooth for the eBOX560-900-FL. It supports the Nvidia JetPack 3.2 SDK, as well as TensorRT, cuDNN, CUDA Toolkit, VisionWorks, GStreamer and OpenCV. The slim and robust box PC is IP40-certified and can be used at temperatures from -10 °C to +50 °C and with vibrations up to 3G. The eBOX560-900-FL has one USB 2.0 port, two Gigabit Ethernet ports, one HDMI 2.0 port and four SMA-type antenna connections. The fanless system also has a 32 GB eMMC and an M.2 SSD PCIe 2.0 x4 socket. The PCI Express mini card slot allows the PC to be connected via 3G/4G/LTE/GPRS, Wi-Fi and Bluetooth.

Axiomtek offers another AI edge computer, the dBOX800, which uses the same hardware basis as the eBOX560-900-FL and therefore differs mainly in the housing and the integrated LTE module.

ICP Germany takes a different approach for its AI IPC and relies on Intel hardware. The Tank-870AI is an inference system with a pre-installed Ubuntu 16.04 LTS operating system and ready-to-use software. The hardware is based on Intel's Skylake or Kaby Lake CPUs and has up to 32 GB of pre-installed RAM. Different AI accelerator cards can be used via the available slots: Mustang-200 is a dual CPU card with Intel's i7/i5 processors. Mustang-F100-A10, on the other hand, uses an FPGA for acceleration and Mustang-V100-MX8 uses Intel's AI processor Movidius.

All of the computer systems described prove that AI applications do not necessarily have to run in a cloud, but can also be executed on site.