NVIDIA Jetson Xavier will arrive in August as a compact and economical solution for Artificial Intelligence

Robert Sole 16 July 2018

0 1.098 2 reading minutes

All the data of the NVIDIA Jetson Xavier is shown, a very compact solution for Artificial Intelligence that will hit the market next August.

Large companies that manufacture chips are entering a blazing race to offer optimized solutions for Artificial Intelligence and Deep Learning. NVIDIA announced Jetson Xavier solutions at CES 2018, which implements the company's most powerful SoC to date. Initially not much data was given, but the company has published all the information of the Jetson Xavier DevKit, which is a compact PC designed for AI, robotics and other specific tasks.

Full details for NVIDIA Jetson Xavier revealed.

The design of this development kit stands out in the first instance. The DevKit implements two modules, one being a special board that offers all the necessary connectivity elements and the other is the development module itself with the powerful SoC. Those who purchase this kit will receive both modules, the cables and a voltage adapter. The main module has dimensions of 100x87x16mm and with the support plate with connectivity, the complete package has dimensions of 105x105mm.

NVIDIA has highlighted that Xavier is manufactured by TSMC using a 12nm process and features a total of 9 billion transistors in a 350mm² die area . Connectivity is based on a 699-pin connector with high-speed I/O ports, offering support for PCIe 4.0. This is NVIDIA's first platform to support PCIe 4.0, reaching transfer speeds of up to 56Gb/s. The company has emphasized that it not only supports the PCIe 4.0 standard but has also been developed to support future I/O standards and is compatible with other Jetson modules.

On the specifications, this Xavier system has a Carmel ARM64 processor manufactured according to the needs and requirements of NVIDIA, which has eight cores for a superscalar architecture. It also offers functional safety, dual execution, parity and ECC thanks to the CPU. The matrix is implemented by a Volta GPU with 512 CUDA Cores. Volta supports FP32, FP16 and INT8 calculations, depending on the needs of the developers. It has a power of 1.3 TFLOPS and has a total of 20 Tensor Cores.