Evaluation of NVIDIA Xavier NX Platform for Real-Time Image Processing for Fusion Diagnostics

Abstract

Real-time image processing is the core component of image plasma diagnostics. Efficient algorithms enable machine protection, contributing to future steady-state operation in nuclear fusion devices. The paper evaluates the applicability of the newest low-power NVIDIA Jetson Xavier NX platform for fusion diagnostics. This embedded NVIDIA Tegra System-on-a-Chip (SoC) integrates a Graphics Processing Unit (GPU) and Central Processing Unit (CPU) on a single chip. General-Purpose computing on Graphics Processing Units (GPGPU) provides high parallelism that is advantageous in image-based calculations. The hardware differences in comparison to the previous NVIDIA Jetson TX2 based on Pascal architecture, including innovations introduced in the Volta architecture for NVIDIA Tegra, are signified. The evaluation is performed on the Wendelstein 7-X (W7-X) stellarator experimental data. Implemented algorithms detect and analyse thermal events in real-time utilising the embedded GPU. Investigated thermal events are strike-lines, overload hotspots, reflections and surface layers. Their detection allows the automated real-time risk evaluation incorporated in the feedback plasma control and interlock systems in the W7-X. The speedup resulting from the upgrade to the Xavier NX platform is presented in the paper, along with techniques pertaining to key hardware differences and programming aspects specific to the NVIDIA Tegra facilitating real-time computing on the low-power embedded device.