Nvidia claims to have the smallest, most powerful AI supercomputer for robotic and embedded computing devices at the edge-- the Jetson Xavier NX, a successor to the Jetson AGX Xavier packed in the form factor of the Jetson Nano.
The Jetson Xavier NX measures 45 x 70mm, making it around the size of a credit card. It is based on the same Xavier SoC as the Jetson AGX Xavier, and and as such has the same Volta GPU and NVDLA accelerator cores. Nvidia says it delivers "server-class" performance of up to 21 TOPS, allowing it to run AI workloads while consuming 15W of power. For lower power consumption it can clock up to 14 TOPS at 10W.
In total, the Jeton Xavier NX module carries 384 CUDA cores and 48 Tensor cores, plus 2x NVDLA. The CPU is a 6-core Carmel Arm 64-bit number with 6MB L2 and 4MB L3 cache, as well as 8GB of LPDDR4x RAM. It supports Ubuntu Linux and the Nvidia JetPack SDK, an AI software stack able to run modern AI networks and accelerated libraries for deep learning, computer vision, computer graphics and multimedia. The hardware is pin-compatible with the Jetson Nano, allowing shared hardware designs and easy upgrades for users of Jetson Nano carrier boards and systems.
The Jetson Xavier NX is available from March 2020. A Jetson GX Xavier Developer Kit with a software patch emulating the Jetson Xavier NX allows developers to start working on compatible applications.
