Linghua ADLINK DLAP Deep Learning Acceleration Platform Product Manual

Enterprise background: Linghua Technology ADLINK

1. * * Industry qualifications and partners**

-NVIDIA Quadro Embedded Partner, Jetson Elite Partner, OEM Preferred Partner; Titanium level member of Intel Internet of Things Alliance.

-We have obtained multiple system certifications such as ISO9001/14001/13485/TL9000 and are listed on the Taiwan Stock Exchange (stock code 6166).

2. * * Global layout**

-Headquarter in Taiwan, China, China, with production lines in Taiwan and mainland; R&D centers are located in the United States, Germany, 

Taiwan, China, and Chinese Mainland; Our global sales and service outlets cover 40+countries across five continents, with over 1800 employees.

3. * * Core business positioning * *: the world’s leading provider of edge computing solutions, 

focusing on embedded and edge AI hardware, focusing on CPU+GPU heterogeneous industrial computing platforms, 

to address the edge AI deployment needs.

Industry background and original intention of DLAP product development

1. Trends in the Edge AI Industry

Traditional mode: devices collect data and send it back to the cloud for analysis;

Industry transformation trend: * * Edge local reasoning, feature matching * *, advantages:

-Low latency: No need to transmit massive data, real-time response;

-High data security: reduce the amount of data transmitted over the network and lower the risk of leakage and tampering;

-Weak network adaptation: Mobile devices can run AI algorithms even without stable networks, adapting to in car and outdoor mobile devices.

2. Inherent pain points of heterogeneous CPU+GPU solutions

The performance of pure CPU or pure GPU single-chip solutions is insufficient, 

but there are three major design challenges in heterogeneous combinations:

1. Higher power consumption; 

2. The device has a larger volume; 

​3. Consumer grade GPUs have a short lifecycle and cannot meet the long supply cycle demands of industry.

3. Core Value of DLAP Platform

The Linghua DLAP series addresses the aforementioned pain points with a focus on balancing SWaP 

(size, weight, power consumption) and AI computing power

-Industrial grade durable design, wide temperature range, anti vibration, wide humidity adaptability to harsh working conditions;

-Choose GPU chips with a long lifecycle to extend the equipment supply cycle;

-Reduce system integrator/OEM/ODM development costs and shorten product launch cycles;

-Provide customized hardware development services and quickly customize edge AI machines based on Nvidia embedded GPU/Jetson modules.

Three major categories of DLAP products (divided by computing power and volume)

The document divides the entire series of products into three tiers: * * low SWaP compact, SWaP performance balanced, 

and high-performance heavy-duty workstations * *, covering the entire scenario from lightweight inference to heavy-duty AI training/multi-channel inference.

（1） Low SWaP Compact (NVIDIA Jetson module with the smallest size and low power consumption)

All equipped with Jetson series SoM modules, no independent desktop graphics card, volume of 0.8~2L, 

FP16 computing power of 0.5~11TFLOPS, focusing on mobile vision, small edge terminals, AI camera NVR, 

supporting fanless wide temperature range of -20 ℃~70 ℃.

Including model:

1. DLAP-201-JT2：Jetson TX2， Volume 0.8L, FP16 computing power 1.5TFLOPS;

2. DLAP-211 series (Nano/JT2/JNX): Unified 0.9L body, equipped with Jetson Nano, TX2NX, Xavier NX respectively;

3. DLAP-301 series (Nano/JNX): 2L body, dedicated for AI NVR, integrated with 8 PoE network ports, used for machine vision camera equipment;

4. DLAP-401 Xavier: equipped with Jetson AGX Xavier, with a volume of 1.8L, FP16 computing power of 5.5~11TFLOPS, and 24V DC power supply.

**General features: dual gigabit Ethernet ports, multiple USB ports, RS232 serial ports, CAN bus, 4-channel DIO, 

M.2/Mini PCIe expansion (supporting 4G/WiFi), SD/mSATA storage, supporting DIN rail, wall mounted, VESA installation, IP40 protection.

（2） SWaP – Performance Balanced Type (DLAP-3000/3100/3200-CF Series, MXM Embedded Solo Display+Intel Core CPU)

Adopting LGA1151 8/9 generation Intel Core i7/i5/i3/Celeron processors, paired with MXM A/B industrial graphics cards 

(P1000/P2000/T1000/RTX3000, up to 120W), with a maximum computing power of 75.2 TFLOPS FP32, 

the compact industrial body is suitable for machine vision, medical imaging, and self-service settlement devices.

Differences among the three sub series:

1. DLAP-3000-CF: Volume 3.2L, 3 Gigabit Ethernet ports, basic expansion;

2. DLAP-3100-CF: Volume 3.2L, upgraded to 5 Gigabit Ethernet ports, built-in TPM2.0, standard audio and DIO;

3. DLAP-3200-CF: volume 5.6L, retains 3 network ports, adds 2 PCIe Gen3 x4 expansion slots, 

can be connected to external devices such as acquisition cards;

**Unified hardware features: up to 64GB DDR4 memory, 6-channel DP display output, M.2 E/B/M multi interface storage/wireless, 12V DC power supply, 

working temperature of 0-50 ℃; The system supports Win10 IoT and Ubuntu 18.04.

（3） High performance heavy-duty platform (DLAP-4000/DLAP-8000, standard PCIe independent RTX graphics card, high computing power)

Targeting heavy AI inference, multi screen rendering, multi-channel visual inspection, supporting single/dual full-length PCIe desktop RTX graphics cards.

1. DLAP-4000 series (9.9L)

-8th/9th generation Core processors, H310 chipset, single PCIe x16 slot, supporting Quadro RTX 4000~8000; 

FP32 has a maximum computing power of 130.5 TFLOPS;

-300W/500W built-in ATX power supply, AC direct supply, maximum 32GB memory, 2 Gigabit Ethernet ports;

2. DLAP-8000 series (15L, flagship industrial GPU workstation)

-9th generation Xeon E2200/Core TE low-power processor, C246 workstation chipset, supports ECC memory;

-Dual full length FHFL PCIe slots, capable of simultaneously carrying 2 RTX8000 and FP32 chips with over 200 TFLOPS computing power;

-4-port hot swappable SATA hard drive (supporting RAID0/1/5/10), 3-channel gigabit Ethernet port, multi-channel serial port/DI/DO, 

TPM 2.0; Supports AC/DC dual power supply, with higher seismic and impact resistance specifications, 

suitable for industrial heavy-duty vision and autonomous driving simulation.

Summary of Hardware General Capability

1. * * Processor scheme dual line parallel**

-ARM route: Full range of NVIDIA Jetson modules (Nano/TX2/Xavier NX/AGX Xavier), low-power fanless;

-X86 route: 8th/9th generation Intel Core, 9th generation Xeon, divided into 35W low-power TE version and 65W standard E version.

2. * * Two forms of graphics GPU**

-MXM Embedded Graphics Card (3000 Series): Compact body, low-power industrial embedded;

-PCIe standard discrete graphics card (4000/8000): ultra-high computing power, supports dual graphics card parallel computing;

-Jetson has built-in integrated GPUs (201/211/301/401 Xavier), which simplify development with integrated modules.

3. * * Enrich I/O and Expansion**

-Network port: 2-8 gigabit channels, 301 series integrated 8-channel PoE power supply network port, compatible with cameras;

-Display: HDMI/DP/DVI, up to 6 synchronous outputs;

-Storage: 2.5-inch SATA, M.2 (B/M/E key), mSATA, SD card, supports RAID;

-Expansion: Mini PCIe, PCIe x4/x8/x16, CAN bus, isolated DIO, serial RS232/422/485, TPM 2.0 secure encryption, eSIM/4G wireless.

4. Environmental adaptability**

-Compact Jetson model: -20 ℃~70 ℃ wide temperature range;

-X86 MXM/PCIe models: standard 0~50 ℃; The flagship 8000 series has stronger earthquake resistance, impact resistance, and ESD protection;

-Humidity of 5%~95% without condensation, certified by EN, UL, CB, FCC, CCC safety regulations and electromagnetic compatibility.

5. * * Power Supply Plan**

-Small edge machine: 12V/24V DC, equipped with an external power adapter;

-High performance workstation: Built in 300/500W AC power supply, some models support 24V DC input, suitable for vehicle/industrial DC scenarios.

Software operating system support

The entire series provides two types of industrial stability systems:

1. Windows: Win10 IoT Enterprise (industrial IoT specific, long-term support version);

2. Linux：Ubuntu 16.04 / 18.04 LTS， Compatible with NVIDIA CUDA and TensorRT inference frameworks.

Landing application scenarios

The document lists mature implementation cases in multiple industries, covering industry, healthcare, retail, transportation and logistics:

1. * * Retail self-service * *: Unmanned cashier canteen, revolving sushi AI pricing system, real-time recognition of goods for quick settlement;

2. * * Medical Imaging * *: Mobile C-arm X-ray machine, endoscope AI image analysis equipment;

3. * * Industrial Automation * *: Robot visual sorting, logistics AGV autonomous navigation (SLAM 3D positioning);

4. * * Special traffic vehicles * *: AI visual collision warning for snow sweepers, identifying pedestrians and obstacles on the road;

5. * * Security Monitoring * *: DLAP-301 PoE NVR, multi-channel camera AI human/object detection;

6. Aerospace, maritime, and autonomous driving: High and low computing power models are respectively adapted to lightweight vehicle terminals 

and ground simulation workstations.

Supporting value-added services

1. * * Deep learning consulting services**

Provide AI performance profiling testing tools, input parameters such as neural networks (AlexNet/MobileNet, etc.), batch size, etc., 

automatically match DLAP hardware, output inference speed, computing power per watt, unit cost computing power comparison, 

and assist customers in selecting and reducing costs.

2. * * Customized hardware development**

Based on long-term research and development experience in embedded modules and carrier boards, 

we quickly customize industry-specific edge AI machines using NVIDIA’s embedded GPU/Jetson modules to meet differentiated SWaP 

and computing power requirements.

3. * * Complete accessory system**

Original factory matching CPU/MXM heat dissipation module, wall bracket, WiFi/4G wireless kit, DC power adapter, various cables, one-stop delivery.