Edge AI ComputerAE Platform32 TOPS RK3588

AE-3588LBT: Fanless Industrial Edge AI Computer | RK3588 32 TOPS

Name: BITECH AE-3588LBT RK3588 Industrial Edge AI Computer
Brand: BITECH
SKU: AE-3588LBT
Availability: InStock

A high-performance, fanless industrial AI box specifically engineered for harsh environments. Powered by the flagship Rockchip RK3588 processor, serving as a robust gateway for machine vision, deep learning inference, and complex industrial automation.

Verify AI & I/O Capabilities Download Datasheet

AE-3588LBT — 32 TOPS RK3588 Industrial Edge AI Box

Engineered for Machine Vision & High-Speed Deep Learning Inference

The AE-3588LBT is a high-performance, fanless industrial AI box specifically engineered for harsh environments. Powered by the flagship Rockchip RK3588 processor, this terminal serves as a robust gateway for machine vision, deep learning inference, and complex industrial automation where reliability and high-speed processing are non-negotiable.

Featuring an extruded aluminum unibody and fanless architecture, the unit operates flawlessly in extreme temperatures ranging from -40°C to +80°C while resisting high-level shock (15G) and vibration. The non-isolated interface design prioritizes low-latency data throughput for high-speed industrial communication.

Designed for 7×24 continuous operation in harsh industrial environments, the AE-3588LBT provides the foundation for long-term industrial AI projects with stable supply chain support.

32 TOPS Scalable Heterogeneous AI Acceleration

Built-in 6.0 TOPS onboard NPU scalable up to 32 TOPS via PCIe M.2 expansion module, supporting INT4/INT8/INT16/FP16 mixed-precision inference.

Ultra-HD Video Pipeline: Simultaneous 4K Capture & 8K Display

HDMI 2.0 Input (4K@60Hz) for real-time capture and HDMI 2.1 Output (8K@60Hz) for ultra-high-definition display, creating a seamless hardware loop for AI-driven visual analysis.

Hardened Fanless Design for Wide-Temperature Environment

Extruded aluminum unibody with fanless architecture operates flawlessly from -40°C to +80°C with 15G shock and high vibration resistance.

High-Bandwidth Automation Interfaces (Cabinet-Optimized Topology)

Dual Gigabit Ethernet, and dual RS232/RS485 ports with non-isolated design prioritizing low-latency data throughput for high-speed industrial communication.

Production-Ready Edge AI & Machine Vision Applications

The AE-3588LBT transforms the raw power of the RK3588 into a production-ready tool for demanding industrial AI and machine vision applications.

Automated Optical Inspection (AOI) & Semiconductor Defect Detection

Perfect for semiconductor wafer inspection where 4K camera streams are captured via HDMI-In, analyzed using 32 TOPS of AI power for microscopic defects, and displayed in 8K for operator review.

Autonomous Mobile Robots (AMR / AGV) Navigation & SLAM

Acts as the primary controller for warehouse robots, utilizing the CAN Bus to manage motor drivers and the NPU for real-time SLAM and obstacle avoidance.

Intelligent Edge Sensing & Smart City Traffic Infrastructure

Deployed in outdoor traffic cabinets to monitor vehicle flow and illegal maneuvers in real-time, relying on the -40°C to +80°C wide-temp range for 24/7 reliability in any climate.

High-End Medical Imaging & Real-Time Surgical Video Processing

Integrated into diagnostic equipment to enhance surgical videos or medical scans in real-time using high-speed AI inference.

Scalable Edge AI Inference & Model Deployment

Built-in 6 TOPS NPU with PCIe M.2 expansion slot for up to 32 TOPS total AI performance.

Deep Learning Model Scaling Performance

Built-in NPU	6 TOPS (INT8 / INT16 / FP16)
AI Expansion	1× PCIe M.2 (M-Key) for NPU accelerator
Maximum Performance	Up to 32 TOPS (6 + 26)
Precision Support	INT4 / INT8 / INT16 / FP16 mixed-precision
Supported Frameworks	TensorFlow, ONNX, Caffe, MXNet, Darknet
AI Toolkit	RKNN-Toolkit2 for model conversion and optimization

AI Performance Scaling

Built-in NPU

6 TOPS

+ M.2 Module

32 TOPS

Supported AI Frameworks & Toolkits

TensorFlowONNXCaffeMXNetDarknet

Deploy current YOLOv8 models today while remaining fully compatible with next-generation deep learning architectures.

Advanced Video Processing Pipeline

Uniquely equipped with HDMI Input and Output for real-time capture and ultra-high-definition display, creating a seamless hardware loop for AI-driven visual analysis.

Display Interfaces

HDMI Output	1× HDMI 2.1 (8K@60Hz output)
HDMI Input	1× HDMI 2.0 (4K@60Hz input)
Additional Display	Supports eDP, LVDS, MIPI display interfaces
Multi-Display	Independent multi-display output supported

Video I/O Capabilities

HDMI 2.0 Input (4K@60Hz) for real-time camera stream capture
HDMI 2.1 Output (8K@60Hz) for ultra-high-definition operator display
Closed-loop system: capture 4K frame grabber or industrial camera streams, AI labeling, output 8K visuals
Critical requirement for semiconductor inspection and high-end medical diagnostics

Low-Latency Industrial I/O & Networking Integration

The AE-3588LBT integrates dual Gigabit Ethernet, high-speed RS232 serial ports, and a 4-bit GPIO block. By deploying a cabinet-optimized, non-isolated hardware topology, this platform deliberately bypasses the propagation delays typically induced by heavy isolation circuitry. This architecture maximizes raw data throughput and guarantees sub-millisecond signal latency, making it the ideal processing node for high-speed local device coordination and vision-guided cabinet automation.

Industrial I/O Interfaces

RS232	2× RS232/485
GPIO	4-bit GPIO
USB	2× USB 3.2 Gen1, 2× USB 2.0

Network & Wireless

2× Gigabit Ethernet (1000Mbps)
WiFi 6 (optional) + Bluetooth 5.2
Mini PCIe for 4G/5G modules with GPS/BeiDou

Cabinet-Optimized Grounding Note

The onboarding RS232 and GPIO interfaces share a unified ground plane with the system’s power input. This non-isolated design is specifically engineered to minimize transmission bottlenecks within a controlled control cabinet or machinery enclosure.
If your system infrastructure requires deployment across long-distance cable runs, faces high electromagnetic interference (EMI), or lacks a unified ground plane, please select the BITECH AE-3588BT, which features native 4000V optical isolation across all industrial I/O pathways.

AE-3588LBT rockchip rk3588 embedded computer

AE-3588LBT RK3588 Edge AI Box I/O Port Layout

Hardened Software Platform & Unattended Field Operations

Designed for 24/7 unattended deployment with comprehensive remote management capabilities.

Open-Root Android & Linux BSP Customization

Native Android 12 with full root access for deep system-level customization and proprietary driver integration.

Secure Over-The-Air (OTA) Fleet Management

Over-the-air system and application upgrades simplify the journey from system integration to global mass-market deployment.

Continuous Remote Connectivity via 4G/5G Modules

Mini PCIe slots for 4G/5G modules with GPS/BeiDou support for reliable remote connectivity.

System Reliability Features

Built-in watchdog (software + hardware)

RTC with backup battery

Fanless, cableless, jumper-less design

Real-time remote monitoring

100% 18-hour power-on burn-in test under load

AE-3588LBT BITECH Edge Computing Platform Evaluation Photos

Real photos to verify ports, mounting, and enclosure design.

AE-3588LBT Industrial Edge AI Computer Front IO

AE-3588LBT Industrial Edge AI Computer Rear IO

AE-3588LBT Industrial Edge AI Computer Overall Dimensions

Request more photos (inside view / labels / packaging)

AE-3588LBT Complete Hardware Specifications

Core Processing & Memory Subsystem

Processor	Rockchip RK3588, Octa-core (4× A76 + 4× A55), up to 2.4GHz
AI Architecture	6 TOPS (onboard) + 26 TOPS (expansion) = 32 TOPS total
Memory	8GB LPDDR4 (options: 16GB / 32GB)
Storage	128GB eMMC (option: 256GB)

Device OS, Firmware & System Stability

Operating System	Android 12 (native, full Root), Debian 11, Ubuntu 20.04
Development Support	Full BSP, kernel source, driver SDKs provided
Stability	Built-in hardware/software Watchdog and RTC with backup battery

Video Pipeline, Physical Interface Layout,Networking, Wireless & Mechanical Dimensions

Video Output	1× HDMI 2.1 (8K@60Hz output)
Video Input	1× HDMI 2.0 (4K@60Hz input)
Additional Display	eDP, LVDS, MIPI DSI
Industrial Bus	2× RS232/485
GPIO	4-bit GPIO
Network	2× GbE LAN (1000Mbps)
Wireless	WiFi 6 (optional), BT 5.2, Mini PCIe for 4G/5G with GPS/BD
Dimensions	148 × 126 × 40mm

This overview highlights decision-level specifications. For full electrical and mechanical details, refer to the datasheet.

Technical Comparison: BITECH AE-3588LBT vs. AE-3588BT

Criterion	AE-3588LBT	AE-3588BT
AI Scalability	6 TOPS core computing power + single PCIe M.2 expansion (up to 32 TOPS)	6 TOPS core computing power + dual PCIe M.2 expansion (up to 58 TOPS)
Network Topology	2× GbE LAN	3× GbE LAN (camera / control / uplink separation)
Industrial Interface Isolation	Industrial interfaces are NOT electrically isolated	4000V optical isolation on all industrial I/O
Deployment Environment	Industrial control automation cabinets and equipment with built-in AOI vision inspection units	ESS sites, substations, harsh field environments

Engineering Note

AE-3588LBT is ideal for cabinet installations where electrical isolation is handled externally. For ESS sites, substations, long cable runs, or environments with high EMI/ground loop risk, choose AE-3588BT with 4000V isolation.

View Full AE-3588LBT vs AE-3588BT Comparison Read: Do You Need Isolated Industrial Interfaces?

OEM & System Customization Support for System Integrators

Designed for system integrators and OEMs who require platform customization and long-term supply chain stability for industrial AI and machine vision projects.

OS customization (Android / Linux / Ubuntu / Debian)

Fieldbus and I/O configuration customization

AI framework and deployment support

Open-root SDKs for deep OS customization

BOM freeze and controlled revision management

Tailored Hardware layout & Deep OS Configuration Services

Requirements Alignment & Scope Definition (1-2 Weeks)

1-2 weeks

I/O, OS, and integration scope definition

Functional Prototype Validation & Sample Evaluation (2-4 Weeks)

2-4 weeks

Hardware validation and software integration

Out-of-Tree Driver Integration & Custom BSP Compilation (4-8 Weeks)

4-8 weeks

BSP, drivers, and system configuration

Hardware-in-the-Loop Quality Testing & Production Ramp-Up (2-4 Weeks)

2-4 weeks

Quality validation and volume production

BITECH Edge Computing Platform Advantages

32 TOPS Scalable AI Headroom

Choosing BITECH AE-3588LBT means investing in a platform with 32 TOPS of scalable AI headroom, allowing you to deploy current YOLOv8 models today while remaining fully compatible with the next generation of complex deep learning architectures without costly hardware swaps.

Simultaneous 4K Input & 8K Output Pipeline

By leveraging BITECH’s unique HDMI Input/Output synergy, engineering teams can easily implement a closed-loop system that captures 4K frame grabber or industrial camera streams for millisecond-level AI labeling and outputs ultra-HD 8K visuals, a critical requirement for semiconductor inspection and high-end medical diagnostics.

Fanless Industrial Reliability with -40°C to +80°C Wide Temperature Operation and 15G Shock Resistance

BITECH’s commitment to a fanless, unibody extruded aluminum design ensures that the AE-3588LBT maintains peak performance under extreme temperature fluctuations from -40°C to +80°C and high-vibration conditions, guaranteeing zero-failure operation in remote or hazardous field deployments.

10 Year Long-Term Supply Commitment with Open-Root SDK and Deep OS Customization Services

As a dedicated B2B partner, BITECH provides a 24-month warranty and a guaranteed 10 year product lifecycle, supported by open-root SDKs and deep OS customization services that simplify the journey from initial system integration to global mass-market deployment.

AE-3588LBT Technical FAQ for Hardware Integration Evaluation

Advanced hardware and software integration details for engineering evaluation.

How do we convert and deploy custom PyTorch or TensorFlow models to the onboard NPU?

Models must be converted to .rknn format using Rockchip's RKNN-Toolkit2, which supports INT4/INT8/INT16/FP16 mixed-precision quantization. The toolkit directly parses standard ONNX, TensorFlow, and PyTorch exports. For optimal inference frame rates on the RK3588, we recommend INT8 asymmetric quantization. This optimization utilizes the full execution pipeline of the NPU, retaining over 98% accuracy for standard object detection models like YOLOv8 and YOLOv10 compared to baseline float32 precision.

Does the single PCIe M.2 slot support simultaneous NVMe SSD storage and an upgraded 26 TOPS NPU accelerator?

No,, the AE-3588LBT has a single PCIe M.2 (M-Key) slot, so it holds either an NVMe SSD or one M.2 NPU accelerator card, not both. If you need high-capacity storage alongside the accelerator, run the OS and data on the onboard 128 GB eMMC (or USB 3.2 storage). For a board with concurrent dual-M.2 architecture, select the AE-3588BT. A note on how the "up to 32 TOPS" figure works: the RK3588's built-in 6 TOPS NPU runs Rockchip's RKNN runtime. An add-on M.2 accelerator (e.g. a Hailo-8 class 26 TOPS module [VERIFY: state the exact module BITECH validates and ships]) is a separate heterogeneous accelerator with its own SDK and model-compilation toolchain — it does not transparently pool with the RK3588 NPU to run a single model. In practice you assign workloads across the two engines (for example, detection on the add-on accelerator, pre/post-processing on the RKNN NPU). The 32 TOPS describes the combined platform headroom across both engines, not a single fused inference budget. Our engineering team provides the model-partitioning reference for your specific pipeline

What is the deterministic glass-to-glass latency for the 4K HDMI 2.0 video input pipeline during live AI inference?

The hardware ingestion latency from the HDMI 2.0 input port to the Rockchip RK3588 system memory via the V4L2 (Video for Linux Two) subsystem is approximately 30 to 45ms at 4K@60fps. Total application-level loop latency depends entirely on your model's computational complexity. For instance, executing a YOLOv8 Nano model under the RKNN runtime adds 8 to 12ms of inference time, yielding a predictable total pipeline latency well under 60ms, which fits automated optical inspection (AOI) trigger windows.

How does the fanless enclosure prevent thermal throttling when running continuous 100% CPU and NPU load at +80°C?

The passive cooling system relies on a heavy-duty, unibody extruded aluminum chassis mated directly to the RK3588 SoC using high-conductivity (6.0 W/m·K) thermal silicon pads. At an absolute ambient temperature of +80°C, the thermal dissipation area keeps the internal silicon junction temperature safely below the 85°C hardware-coded throttling threshold during typical machine vision duty cycles. For sustained maximum computational load at absolute temperature extremes, a well-ventilated enclosure or cabinet-level extraction fan is recommended.

How should grounding and electrical protection be handled for the non-isolated RS232 and GPIO interfaces inside the cabinet?

Because the onboard RS232 serial ports and 4-bit GPIO lines share a common ground with the system power input, a single-point equipotential ground network must be established across the entire control cabinet.

To ensure signal integrity and protect the Rockchip RK3588 processor from stray transient voltage in controlled environments:

Common Grounding: Always connect the signal ground (GND) of the peripheral device directly to the AE-3588LBT’s ground reference to eliminate potential discrepancies.
Cable Routing: Keep serial and I/O lines separated from high-voltage AC cables or heavy inductive loads (such as variable frequency drives or relays) to reduce electromagnetic interference (EMI).
External Isolation: If your deployment requires long cable runs (greater than 10 meters) or connects to field devices prone to severe ground loops, you must install external inline opto-isolators.

For infrastructure lacking a unified ground plane or requiring native 4000V optical isolation directly on the I/O block, we strongly recommend deploying the AE-3588BT instead.

Is the Linux BSP (Ubuntu 20.04 / Debian 11) supplied with upstream kernel source for compiling proprietary kernel modules?

Yes, we provide full development access. The AE-3588LBT ships with an open-root BSP based on Linux Kernel 5.10 / 6.1 LTS. BITECH provides the complete kernel source code, cross-compilation toolchains, and default defconfig profiles. Your engineering team can freely compile custom out-of-tree kernel modules, patch V4L2 capture structures for proprietary industrial cameras, or integrate custom USB/Serial fieldbus hardware abstraction layers at the root level.

Evaluate the AE-3588LBT

Tell us your enclosure and I/O — we confirm this is the right RK3588 for it.

Cabinet-mounted with a unified ground plane? The non-isolated AE-3588LBT gives you lower-latency I/O at lower cost. Send us your camera pipeline, NPU target, and OS, and an engineer returns a configuration and sample terms — not a brochure.

NPU scaling plan: 6 TOPS onboard or +M.2 accelerator to 32 TOPS
4K HDMI-in / 8K-out loop

Where AE-3588LBT Is Deployed