Specifications

Model Numbers

0802S, 0802, 0803, 0804, 0804S, 0805S, 0806S, 0807S, 0808S

AI Performance

52, 52, 78, 104, 104, 130, 156, 182, 208 TOPS

Architecture

Multiple Hailo-8 AI processors per card

Form Factor

Single slot, HHHL or HHFL; 68.5x130mm / 68.9x167.65mm / 68.9x254mm

Interface

PCIe x8; PCIe x16 with bifurcation; PCIe x16

Max TDP

15W-45W (lower models) · 45-75W (higher models)

Power Efficiency

400 FPS/W on ResNet-50

Analytics Streams

32 basic / 16 advanced up to 128 basic / 64 advanced

Temperature

Industrial -40°C to 85°C

OS / Frameworks

Linux, Windows 10/11 · TensorFlow, TFLite, Keras, PyTorch, ONNX

Pricing

From $249 (entry models); higher models via Hailo partners

Ecosystem

Aetina AI-MXM-H84A, Lanner Falcon H-8, MiTAC MiAi-H8, Sunix AIEH1000, Advantech EAI-3300

Overview

The Hailo-8 Century is a family of PCIe accelerator cards that scale edge AI inference from 52 TOPS to 208 TOPS by aggregating multiple Hailo-8 AI processors on a single card. Nine model numbers (0802S through 0808S) cover three mechanical footprints and two thermal design points, from a 68.5 x 130mm single-slot HHHL card drawing 15-45W up to a 68.9 x 254mm card drawing 45-75W. Cards run from either a PCIe x8 slot or a PCIe x16 slot (with bifurcation on some variants) and are rated for the full industrial -40°C to 85°C range. Hailo positions the Century line against datacenter-derived GPU cards for video analytics on the strength of power and cost efficiency — 400 FPS/W on ResNet-50 and entry pricing from $249.

Key Benefits

Scale channels, not power: each Century card supports 32 basic / 16 advanced analytics streams at the entry point and up to 128 basic / 64 advanced streams at 208 TOPS, all within a 75W ceiling. That lets a standard 2U or 4U industrial server handle a large camera estate without the power, cooling and cost overhead of datacenter GPUs. Deterministic multi-model inference: the Hailo dataflow architecture runs many models and many streams concurrently with low latency, which suits real-time video analytics rather than batch throughput. Deployable anywhere: HHHL and HHFL single-slot cards fit industrial, telecom and rugged chassis; -40°C to 85°C operation removes the need for a climate-controlled rack. Proven integration path: partner products from Aetina, Lanner, MiTAC, Sunix and Advantech already wrap Hailo-8 silicon into MXM modules, edge systems and motherboards.

Applications

High-density video management and analytics servers, smart city and traffic monitoring, retail and campus loss prevention, industrial safety and PPE monitoring, NVR appliances, and adding edge inference capacity to existing x86 servers via a spare PCIe slot.

Request a Quote — HAILO-8 CENTURY — 52-208 TOPS PCIE EDGE AI INFERENCE CARD

QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.

Get Your Quote →

Related Products

NVIDIA L4 — 24GB Ada Lovelace Inference GPU Axelera Metis AIPU — Edge AI Inference Accelerator Hailo-8 M.2 AI Acceleration Module — 26 TOPS