
Edge Computing Hardware Guide: Gateways, Industrial PCs, Embedded Controllers

Edge Computing Hardware Guide: Gateways, Industrial PCs, and Embedded Controllers
Edge computing brings processing closer to where data is generated — reducing latency, bandwidth costs, and cloud dependency. Here's how to choose edge computing hardware, from simple gateways to AI-capable industrial PCs.
Edge Hardware Tiers
| Tier | Processing | Memory | Storage | OS | Cost | Example |
| Light gateway | Cortex-A7 | 256MB-1GB | 4-8GB eMMC | Linux/RTOS | $30-80 | Raspberry Pi CM4 industrial |
| Mid-range | Cortex-A53/A72 | 1-4GB | 8-32GB | Embedded Linux | $80-250 | NXP i.MX 8M Plus |
| Performance | Cortex-A78/X1 | 4-16GB | 32-256GB | Linux/Windows IoT | $200-800 | NVIDIA Jetson Orin NX |
| Industrial PC | x86 (Atom/Core) | 8-64GB | 128GB-2TB SSD | Windows/Linux | $500-3000 | Advantech, Kontron |
Light Gateways: Protocol Translation + Basic Filtering
Best for: Modbus-to-MQTT bridging, basic data filtering, simple dashboards
Don't use for: Video processing, ML inference, heavy database workloads
Popular hardware:
- Raspberry Pi Compute Module 4 (industrial variant, -20°C to 85°C)
- STM32MP157 (Cortex-A7 + Cortex-M4, real-time capable)
- Mediatek Genio 350 (Yocto Linux, 10-year availability)
Mid-Range: Local Analytics + Dashboard
Best for: Video analytics (single camera), industrial protocol gateways, local HMI
Popular hardware:
- NXP i.MX 8M Plus (integrated 2.3 TOPS NPU)
- TI AM62x (Linux, industrial temperature, 10+ year availability)
- Rockchip RK3568 (quad Cortex-A55, NPU, under $30 in volume)
Chinese options:
- Rockchip RK3588 (quad A76 + quad A55, 6 TOPS NPU, 8K decode)
- Allwinner T527 (octa-core A55, 2 TOPS NPU, cost-optimized)
Performance Tier: AI Inference at the Edge
Best for: Multi-camera video analytics, defect detection, predictive maintenance
Popular hardware:
- NVIDIA Jetson Orin NX (70-100 TOPS, CUDA ecosystem)
- Hailo-8 M.2 module (26 TOPS, PCIe, pairs with any x86/ARM host)
- Google Coral M.2 (4 TOPS, TensorFlow Lite)
Industrial PCs: When You Need x86
Best for: Legacy Windows software, complex HMIs, factory floor integration
Key brands:
- Advantech (UNO, ARK, MIC series — wide temperature)
- Kontron (KBox series, railway certified)
- OnLogic (fanless, industrial, customizable)
Chinese IPC makers:
- EVOC (研祥) — China's largest IPC manufacturer
- Norco (华北工控) — Competitive pricing, wide range
- AAEON (研扬) — Asus subsidiary, good quality
Selection Checklist
- Temperature range: 0-50°C (commercial) vs -40 to 85°C (industrial). Industrial-rated hardware costs 2-3x more.
- Mounting: DIN rail for control cabinets, VESA for displays, wall-mount for gateways
- I/O count: 2x Ethernet for network segmentation, isolated RS-485 for industrial protocols, CAN for vehicle interfaces
- Certifications: CE/FCC for general use, EN 50155 for railway, ATEX for hazardous areas
- Lifecycle: Industrial suppliers guarantee 7-15 year availability. Consumer-grade SBCs change every 1-2 years.
- Power: 9-36V DC input for industrial/vehicle use. Don't rely on USB power in the field.
Sourcing Edge Hardware
- Single-board computers: Lead times have normalized (4-8 weeks for industrial variants). CM4 shortages (2021-2023) are resolved.
- NVIDIA Jetson: Orin series is in production; stock available through Arrow and Avnet.
- Custom carrier boards: For volume >1000 units, design your own carrier around a compute module. Costs $10-30K for design + $50-100/board at volume.
Related Articles
- AI Chips for Edge Computing: A Buyer's Guide — NPUs, TPUs, inference accelerators
- FPGA vs ASIC vs GPU for AI Acceleration — AI hardware paths compared
- IoT Module Selection Guide — Wireless connectivity for edge gateways
Need help sourcing these components?
PartsCube Global stocks all alternatives mentioned in this guide. Search our catalog or submit your BOM for a quote.
Chat on WhatsApp