Neousys PCIe-NX154-JON16 Jetson Orin NX 16GB Intelligent Frame Grabber Card with 4x PoE+ 2.5GbE

Adding deep learning to an existing vision system usually means a new computer or a GPU the chassis cannot take. The PCIe-NX154-JON16 puts the AI inside a PCIe slot instead. It is the 16GB build in the Neousys PCIe-NX154 series, an AI frame grabber card that carries its own NVIDIA Jetson Orin NX 16GB module and powers four GigE or IP cameras directly.

Four 2.5GBASE-T RJ45 ports connect the cameras, and the datasheet names the Intel I225-IT for these links. The card talks to the host over Gen2 PCI Express x1 at 2.5 Gb/s total bandwidth, fitted in a single PCIe x4 slot; Neousys describes this as 2.5GbE host communication. Because image capture, video streaming, pre-processing and inference all run on the card, deep-learning work is taken off the host CPU and GPU. Neousys names 19-inch rack-mount IPCs and off-the-shelf box PCs as typical hosts.

Each port supplies up to 25.5 W under IEEE 802.3at PoE+, with a 50 W total budget across the four ports, which is worth checking against camera draw; our white paper on PoE power budgeting for multi-camera edge AI walks through the sums. The module has 16GB LPDDR5 at 3200 MHz, and Neousys rates the card at 100 INT8 TOPS from 1024 CUDA cores, 32 Tensor cores and 2 NVDLA engines, with up to four 4K30 decode streams. The PCIe-NX154-JON8 uses the 8GB module, and the PCIe-NX156-JON16 swaps the PoE ports for six USB 3.2 ports.

The card operates from -25 to 60 °C with airflow in 20W TDP mode without throttling at 60 °C; 2.5G links are limited to 60 °C, and a wide-temperature NVMe is needed outside 0 to 60 °C. It measures 167.7 x 111 mm and weighs 0.4 kg. A 12 V DC input serves standalone development or builds drawing more than 66 W in total. RS-232 and isolated RS-485 ports sit on the card, with development Gigabit Ethernet, USB 2.0, micro USB OTG and 4K60 DisplayPort. It works with Windows and Linux hosts and carries CE and FCC Class A to EN 55032 and EN 55035.