With the Jetson Xavier NX Nvidia has introduced a new developer board for the machine learning sector. The board is to be used, for example, in robots and in edge computing.
Since 2014, Nvidia offers under the brand name Jetson various development boards for deep learning use. With the Jetson Xavier NX, the manufacturer has introduced the latest model of the product line. Jetson Xavier NX is designed as a plug-in module whose footprint is slightly smaller than that of a credit card. The board is pin-compatible with the Jetson Nano and should therefore be well suited for upgrades.
With a power of 15 watts, it should bring the board to 21 TeraOPS (INT8). In 10-Watt mode, there are still 14 TeraOPS. The performance in deep learning benchmark MLPerf Interference should be about 15 times higher in the Jetson Xavier NX than in the 2017 introduced Jetson TX2, which was even larger.
Jetson Xavier NX: An overview of the features of the machine learning board
Jetson Xavier NX has a 64-bit hex ARM CPU called Carmel. These are joined by 384 Cuda graphics cores, 48 Tensor cores and two NVDLA deep learning accelerators. CPU and GPU timing will vary depending on whether the board is operating in 10-watt or 15-watt mode. The board has 8 gigabytes of LPDDR4x RAM and 16 gigabytes of eMMC memory.
The selling price should be at 399 US dollars. This means the hardware is much more expensive than the Jetson Nano, which is offered for $ 99. However, it also brings significantly less power.