Mobilint ARIES MLA400 PCIe Card — 320 TOPS Quad-ARIES Inference Accelerator

Specifications

Overview

The MLA400 packs four ARIES NPUs onto a single PCIe Gen5 x16 card, producing 320 TOPS of AI inference within a 135 W power budget. With 64 GB of onboard LPDDR4X running at up to 266.8 GB/s, the card is dimensioned for multi-stream video analytics, multi-user LLM inference and multimodal workloads that would otherwise require several conventional GPU cards and far more rack power.

The card is 268 mm long, 112 mm high and 19 mm thick, and draws from both the PCIe slot and an auxiliary 8-pin connector. Mobilint offers both a passive slim cooling design for high-airflow servers and a heatsink-plus-active-fan variant for chassis where directed airflow is limited, with a stated deployment profile of sustained throughput in ventilated environments.

Software support mirrors the rest of the ARIES family: the SDK qb driver and runtime, Linux and Windows operating systems, and direct compilation of PyTorch, TensorFlow, TFLite, ONNX and Keras models.

Key Benefits

320 TOPS in a single slot-class card for density-sensitive inference racks. 266.8 GB/s memory bandwidth keeps large vision and language models fed. Passive or active cooling to match server airflow. One software stack across PCIe, MXM and box-level ARIES products.

Applications

Rack-scale video analytics and smart-city inference; on-premises multi-user LLM serving; multimodal document and video understanding; factory-wide visual quality inspection; telecom and broadcast content inference; sovereign and private-cloud AI deployment.

Related Products