Supermicro has created a compact fan-less AI edge computer for manufacturing and retail, built around Nvidia’s Jetson Orin NX processor.

Supermicro ARS-E103-JONX-H2 AI edge computer

For inferencing it is “capable of delivering up to 157Top/s for multiple concurrent AI pipelines, combined with high-speed IO interface options including 10Gbit/s Ethernet, 5G and Wi-Fi”, according to Supermicro.

Dubbed ARS-E103-JONX, it measures : 185 x 80 x 140mm and works over -25 to +60°C with 0.7m/s airflow.

Although Supermicro claims 157Top/s, Nvidia rates the Jetson Orin NX at up to 100 sparse INT8 Top/s and 50 dense INT8 Top/s for the version with 16Gbyte of on-board EEC LPDDR5 memory (’16G’), or up to 70 sparse and 35 dense INT8 Top/s for the 8Gbyte (‘8G’) version.

Application processing in the comes from 64bit Arm Cortex-A78AE (v8.2) cores, eight in the 16G version of Orin NX and six in 8G version.

ARS-E103-JONX-H2 has the eight core CPU, which means that it ‘Ampere’ GPU has 1,024 CUDA cores and 32 Tensor cores, running at 918MHz.

Supermicro ARS-E103-JONX-H2 AI edge computer annotated

On top of this are two 614MHz 20Top/s (sparse INT8) NVDLA deep learning accelerators.

ARS-E103-JONX-L2 is the six core 8Gbyte version, which has a slower (765MHz) GPU and only one (610MHz) NVDLA.

For main memory there is a M.2 PCIe 4.0 (x4) NVMe slot (M-key 2280 default).

Interfaces include 1x 10Gbit/s Ethernet, 4x 1Gbit/s Ethernet, 1x USB 3.2 Gen2 Type-C, 3x USB 3.2 Gen2 Type-A, HDMI for a display, 2x RS232/422/485 serial, 1x CAN, 4x digital inputs and 4x digital outputs.

Options include a DIN rail clip for mounting.

Find the ARS-E103-JONX-H2 on this Supermicro web page

Announced very recently, Nexcom picked Nvidia’s Jetson Orin Nano for an edge AI PC for smart agriculture

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved