T-Chip Group
Intelligent hardware website
Innovate intelligent hardware and solutions.
Server Business Website
Server products and efficient solutions.
Consumer Business Website
StationPC NAS products and services.

CSD2-N128 AI Server

128 Compute Nodes in Total

· Mainstream SoC Platform Support

· Fully Hot-Swappable Modular Design

· Android Multi-Instance Support

· 80 Gbps Aggregate Peak Bandwidth

· Granular Network Security Control

· Smaller Single-Point Failure Impact Range

· Up to 3840-Channel AI Video Processing

· Highly Integrated Server Design


Manual Wiki Download

Gemma

Llama

Qwen

Stable Diffusion

CSD2-N128
CSD2-N128 AI Computing Server

CSD2-N128 features 16 built-in compute blades (128 compute nodes in total), with each node delivering 6-60 TOPS of computing power. Platform options include Qualcomm, Rockchip, Sophgo and SpacemiT. It supports private deployment of mainstream AI large models and multiple deep learning frameworks. Equipped with 8 x 10GbE ports, achieving peak switch bandwidth of 80 Gbps. With standard 2U rackmount server chassis design, it also comes with intelligent BMC management system.

Mainstream SoC Platform Support

CSD2-N128 offers a wide range of AI processor options, including mainstream hardware platforms such as Qualcomm, Rockchip, SOPHGO, and SpacemiT, adapting to diverse computing scenarios.

All-Module Hot-Swap Design

The server adopts fully hot-swappable modular architecture, integrating BMC network modules, switch network modules, power modules, and 16 compute blade modules. Each compute blade is standardly equipped with 8 independent compute nodes, enabling high-density deployment of up to 128 compute nodes.

Android Multi-Instance Support

Adopts Android container-level isolated multi-instance architecture. Through lightweight virtualization scheduling mechanism, it efficiently reuses SoC hardware resources, significantly improving chip computing power and overall system resource utilization.

80 Gbps Aggregate Peak Bandwidth

Equipped with 8 x 10GbE SFP+ ports, delivering an aggregate peak bandwidth of up to 80 Gbps to meet high-bandwidth application requirements. An independent BMC management network interface separates the management network from the business network, ensuring secure and reliable network communication.

Granular Network Security Control

Equipped with 8 x 10GbE SFP+ network ports for high-speed, stable communication to meet the needs of various application scenarios. An independent BMC management network interface separates the management network from the business network, ensuring security and reliability of network communication.

Smaller Failure Impact Range

Each compute blade module is equipped with 8 core boards. When a single core board fails, maintenance affects only its corresponding 8 compute nodes — significantly better than the industry norm of 16-20 affected nodes.

3840-Channel AI Video Processing

Supports AI processing for up to 3,840 video streams (actual processing performance may vary depending on SoM specifications and type of AI model being run). With multi-task concurrent processing capabilities, it is widely applicable in AI application scenarios such as intelligent security and edge computing.

Efficient and Low-Cost

Highly integrates computing units, storage, USB controllers, network controllers, power management controllers, and sensors into a single system. It provides an all-in-one SDK for deep learning development, along with a suite of software tools including low-level drivers, compilers and inference deployment tools, reducing users' procurement, development and operational costs.
128 Computing Nodes, Powerful Performance

Fully configured system can deploy up to 128 compute nodes, with each node delivering 6-60 TOPS of computing power. Each node can support 5-10 containers deployed in parallel based on business requirements, enabling the entire system to virtualize and host 640-1,280 system containers. This effectively revitalizes hardware resources and significantly improves overall resource utilization efficiency.

Application Scenarios

Widely applicable in industry fields such as Edge Computing, Large Model Localization, Smart City, Smart Healthcare, Smart Industry and Intelligent Security.

Edge Computing
Edge Computing
Private Deployment of Large Models
Private Deployment of Large Models
Smart City
Smart City
Smart Healthcare
Smart Healthcare
Smart Industry
Smart Industry
Intelligent Security
Intelligent Security
Interfaces
Specifications

CSD2-N128Q8550

CSD2-N128R3588S

CSD2-N128R3576

CSD2-N128SPK3

Technical Specifications

Launch Status

Launch in June 2026

Server form

2U rack-mounted computing power server

Architecture

ARM architecture

RISC-V architecture

Number of nodes

16 compute blades (128 distributed compute nodes) + 1 control node

Compute nodes

Octa-core 64-bit processor Qualcomm QCS8550, up to 3.36GHz

Octa-core 64-bit processor RK3588S, up to 2.4GHz

Octa-core 64-bit processor RK3576, up to 2.2GHz

Octa-core 64-bit processor SpacemiT Key Stone K3, up to 2.4GHz

Video encoding

8K@30fps/4K@120fps H.264

H.264: 1×8K@30fps, 16×1080P@30fps

H.264: 1×4K@60fps

4K@60fps H.264

Video decoding

8K@60fps/4K@240fps H.264/VP9/AV1

8K@60fps/4K@120fps (VP9/AVS2) 8K@30fps (H.264/AVC/MVC) 30×1080P@30fps (H.264)

1×4K@120fps (VP9,AVS2,AV1) 1×4K@60fps (H.264/AVC)

4K@120fps H.264/VP9

Control nodes

Octa-core 64-bit processor RK3588, main frequency up to 2.4GHz, the highest computing power is 6TOPS

AI computing power

6144TOPS (48T × 128, INT8)

768TOPS (6T × 128, INT8)

7680TOPS (60T× 128, INT8)

RAM

16GB LPDDR5X × 128

16GB LPDDR5 × 128 (4/8/16/32GB)

8GB LPDDR4/LPDDR5 × 128 (4/8/16GB)

32GB LPDDR5 × 128 (8/16/32GB)

Storage

256GB UFS4.0 × 128

256GB eMMC × 128 (16/32/64/128/256GB)

64GB eMMC × 128 (16/32/64/128/256GB)

128GB UFS2.2 × 128

Power

2 × 1300W hot-swappable power supplies, 1+1 redundancy support

Fan module

14 high-speed cooling fans

Physical Specifications

Size

Standard 2U rack servers: 495.60mm × 928.52mm × 88.80mm

Installation requirements

IEC 297 Universal Cabinet Installation: 19 inches wide and 800 mm deep and above Retractable slideway installation: The distance between the front and rear holes of the cabinet is 543.5mm~848.5mm

Environment

Operating Temperature: 0ºC ~ 30ºC, Storage Temperature: -40ºC ~ 60ºC, Operating Humidity: 5% ~ 80%RH (non-condensing)

Software Specifications

BMC

The BMC management system is integrated with the web-based management interface, supporting Redfish, VNC, NTP, monitoring advanced and virtual media, and the BMC management system can be redeveloped

Large language models

All models support private deployment of ultra-large-scale parameter models under the Transformer architecture, such as large language models including Deepseek-R1 Series, Gemma Series, Llama Series, ChatGLM Series, Qwen Series, Phi Series, etc.

Visual large model

K3: Supports private deployment of all vision large models QCS8550: Supports private deployment of vision large models including Qwen2.5-VL, InternVL3, etc.

AI Painting

K3: Supports private deployment of all image generation models QCS8550: Supports private deployment of the Stable Diffusion image generation model

Deep learning

All models: Support traditional network architectures such as CNN, RNN, LSTM, and support various deep learning frameworks such as TensorFlow, PyTorch, PaddlePaddle, ONNX, and Caffe. Support custom operator development and Docker containerization management technology

Interface Specifications

Internet

8 × 10Gbps SFP+, 1 × Gigabit Ethernet (RJ45, MGMT is used as BMC management network)

Console

1 × Console (RJ45, BMC debug serial port, baud rate 115200)

Display

1 × VGA (maximum resolution 1080P, BMC management display)

USB

3 × USB3.0, 1 × Type-C (OTG)

Button

1 × Power, 1 × UID, 1 × Recovery, 1 × Reset
Customization

Firefly team, with over 20 years of experience in product design, research and development, and production, provides you with services such as hardware, software, complete machine customization, and OEM server.

Hardware Solution

Software Solution

Turnkey Solution