Product Overview

Product Overview#

The Mobilint ARIES NPU

ARIES is Mobilint’s AI accelerator NPU — engineered for high-performance, low-power deep-learning inference at the edge. It powers a family of products, from PCIe add-in cards to a fully integrated edge AI PC. This manual covers the ARIES hardware line-up and how to run it on a host system: identify your product, install the driver, keep firmware up to date, and tune the accelerator’s cores for your workload.

  • 80 TOPS — AI inference performance

  • 25 W — Typical power consumption

  • 3,082 FPS — ResNet-50 inference

  • 11.3 TPS/User — Llama-3.2-3B-Instruct

Model

Specification

MobileNetV 2

11,551 FPS

ResNet-50

3,082 FPS

YOLO11s

784 FPS

YOLO11l

274 FPS

YOLO26m

343 FPS

Llama-3.2-3B-Instruct

14.62 TPS/User

Qwen3-1.7B

25.92 TPS/User

Installation#

Prepare a host system so it can detect and use an ARIES device.

  • Driver Installation: install the kernel driver that exposes the NPU to your operating system.

  • Firmware Update: bring a PCIe device up to the latest firmware for stable, optimal operation.

  • Checking Compatibility: verify your driver, runtime, and firmware versions are a supported combination.

Guide#

Get the most out of the accelerator once it is up and running.

  • Core Mode: the four core-collaboration modes — Single, Multi, Global4, and Global8 — and how to choose the right one for your workload.

  • ModelConfig Configuration: assign the NPU cores and core mode for each model through the runtime’s ModelConfig object.

Supported Hardware#

ARIES ships in several form factors, from a PCIe add-in card to a fully integrated edge AI PC. Select your product to review its host-system and software prerequisites.

  • MLA100 PCIe Card: ARIES NPU on a PCIe add-in card for servers and workstations.

  • MLA100 MXM Module: ARIES NPU as an MXM 3.1 module for embedded and carrier-board systems.

  • MLX-A1 Edge AI Box: Turnkey, self-contained edge AI Box with an ARIES MXM module preinstalled.

  • MLA400 PCIe Card: Featuring a quad-ARIES architecture, the MLA400 PCIe card delivers massive AI throughput with outstanding power efficiency. Documentation coming soon.