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.

CSA1-N8 AI Server

Up to 256T Computing Power

· Private deployment of AI models  

· Multiple deep learning frameworks  

· 256-channel video AI processing capability

· Supports 3.5-inch SATA3.0 HDD/SSD  

· Dual 10-Gigabit SFP+ ports ＆ Gigabit Ethernet

· Standard 1U rack server size

· Includes aBMC management system

· Highly integrated server design

Manual Wiki Download

Llama

Qwen

Stable Diffusion

CSA1-N8
CSA1-N8 AI Server

CSA1-N8 features built-in distributed computing nodes with 8 computing platforms, each capable of delivering up to 32 TOPS of computing power. It supports the private deployment of mainstream AI large models and multiple deep learning frameworks. Equipped with two 10-gigabit Ethernet ports and a gigabit Ethernet port, it also supports expandable SATA 3.0 hard drives.

Private Deployment

Supports private deployment of mainstream large language models (such as Gemma, Llama, Qwen, etc.), vision large models (such as EfficientVIT, SAM, TAM, etc.), image generation models (such as Flux, Stable Diffusion, etc.), and other large-scale models.

Deep Learning Frameworks

Supports traditional network architectures (CNN, RNN, LSTM) and mainstream deep learning frameworks (TensorFlow, PyTorch, PaddlePaddle, ONNX), with additional capabilities for custom operator development and Docker containerization.

Video AI Processing Capability

Capable of AI processing for 256 concurrent video streams, featuring robust multi-tasking performance. Actual throughput depends on the core board and AI model, enabling diverse applications in intelligent security and edge computing.

3.5-inch SATA 3.0 HDD/SSD

Configured with one 3.5-inch (or 2.5-inch) drive bay that supports SATA 3.0 HDD/SSD expansion, the device can be easily scaled to terabyte-level capacity. It supports hot-swapping for fast drive replacement, meeting the one-stop deployment requirements for file management, data backup, video surveillance, and other applications.

Dual 10-Gigabit SFP+ ports

Featuring dual 10-Gigabit SFP+ and dual Gigabit Ethernet ports, it delivers the high-speed, stable connectivity required for high-bandwidth application scenarios.

Standard 1U Rack Server Size

Features a standard 1U rack-mounted chassis design with flexible deployment capabilities, fully compatible with most data center cabinet types. The robust construction incorporates 6 high-efficiency cooling fans to ensure 7×24 continuous stable operation.

aBMC Management System

Equipped with an advanced aBMC intelligent management system that enables real-time monitoring, software/hardware configuration, troubleshooting, anomaly alerts, system upgrades, and remote maintenance. Supports secondary development for customized requirements.

Efficient and Low-Cost

Server highly integrates computing units, storage, universal serial bus, network controller, power controller, and sensors into one, providing SDK all-in-one deep learning development toolkit, underlying driver environment, compiler, inference deployment tool, and a series of software tools to reduce users' procurement, development, and operation costs.
256 TOPS Computing Power NPU

CSA1-N8 features eight distributed computing nodes, configurable with either Sophon BM1684X or BM1684 chips. Each node delivers up to 32 TOPS or 17.6 TOPS of computing power, respectively. The number of nodes is also customizable (up to eight), providing robust computational support for AI and deep learning applications.

Interfaces
Specifications

CSA1-N8S1684X

CSA1-N8S1684

Technical Specifications

Server form

1U rack-mounted computing power server

Architecture

ARM architecture

Number of nodes

8 distributed computing nodes + 1 control node

Compute nodes

Octa-core 64-bit processor BM1684X, up to 2.3GHz

Octa-core 64-bit processor BM1684, up to 2.3GHz

Video encoding

H.264: 3×4K@25fps, 12×1080P@25fps

H.264: 2×1080P@25fps

Video decoding

H.264: 8×4K@25fps, 32×1080P@25fps

H.264: 32×1080P@30fps

Control nodes

Octa-core 64-bit processor RK3588S, main frequency up to 2.4GHz, the highest computing power is 6TOPS

AI computing power

256TOPS (32T × 8, INT8)

140.8TOPS (17.6T × 8, INT8)

RAM

16GB LPDDR4/LPDDR4X × 8

12GB LPDDR4/LPDDR4X × 8 (12GB/16GB Optional)

Storage

64GB eMMC × 8 (64GB/128GB Optional)

32GB eMMC × 8 (32GB/256GB Optional)

Storage expansion

3.5-inch/2.5-inch SATA3.0 SSD hard drive slot × 1 (Supports hot swapping; BMC can directly operate the hard drive, and computing child nodes can indirectly access the hard drive through the network sharing method provided by BMC)

Power consumption

Normal: 300W, Max: 430W

Normal: 170W, Max: 350W

Fan module

6 high-speed cooling fans

Physical Specifications

Size

490.0mm(L) × 417.3mm(W) × 44.4mm(H)

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

Full weight

Server net weight: 6.7kg, total weight with packaging: 8.9kg

Environment

Operating Temperature: 0ºC ~ 42ºC, Storage Temperature: -40ºC ~ 60ºC, Operating Humidity: 20% ~ 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

BM1684X: Support the privatization of ultra-large-scale parametric models under the Transformer architecture, such as Deepseek-R1 series, Gemma series, Llama series, ChatGLM series, Qwen series, Phi series and other large language models

Visual large model

BM1684X: Support the privatization deployment of large visual models such as ViT, Grounding DINO, SAM, etc.

AI Painting

BM1684X: Support the private deployment of Flux, Stable Diffusion, and Stable Diffusion XL image generation models

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

2 × 10G Ethernet (SFP+), 2 × Gigabit Ethernet (RJ45, 1 management network port, 1 ordinary network port)

Display

1 × HDMI2.0 (Maximum resolution 1080P, main processor core board display)

USB

2 × USB3.0 HOST, 1 × Type-C (USB3.0 OTG, processor core board debugging)

Others

1 × SIM Card, 2 × 4G antenna
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