Nvidia's embedded compute arm has announced the latest Jetson computer-on-module (COM) design, dubbed the Jetson AGX Xavier, which it claims can offer 32 trillion operations per second (TOPS) of compute performance in as little as 10 watts.
Nvidia's efforts in the high-performance embedded compute market began back in 2014 with the launch of the Jetson K1 single-board computer (SBC), a high-priced device with which it aimed to encourage both hobbyists and professionals to adopt its Tegra system-on-chip (SoC) hardware and its combined Arm CPU and in-house GPU processing cores. Its replacement, 2015's Jetson TX1, switched to a computer-on-module (COM) format, while 2017's Jetson TX2 switched to the Tegra X2 Parker SoC and enjoyed a corresponding increase in performance.
This year's big launch is the Jetson AGX Xavier, with which Nvidia is targeting embedded artificial intelligence (AI) and robotics applications with the promise of 32 trillion operations per second (TOPS) in a 10W power envelope - 'about as much', the company crows, 'as a clock radio'.
The hardware itself is based around an eight-core Arm v8.2 64-bit CPU with 8MB of L2 and 4MB of L3 cache memory, a 512-core Volta-based graphics processing unit (GPU) with Tensor Core AI-focused acceleration hardware added in, 16GB of LPDDR4x memory on a 256-bit bus with 137GB/s throughput, along with 32GB of eMMC storage. Additional compute acceleration is provided via two Nvidia Deep Learning Accelerators (NVDLAs) and a seven-way vision processor linked to a pair of codecs supporting 4Kp60 operation.
More information on the Jetson AGX Xavier is available on the official website, while pricing has been set at $1,399 (around £1,106 excluding taxes).