Firefly’s CSB1-N10 series AI cluster servers can deliver up to 1000 TOPS of AI power with Rockchip or NVIDIA Jetson Modules

Firefly has recently introduced the CSB1-N10 series AI cluster servers designed for applications such as natural language processing, robotics, and image generation. These 1U rack-mounted servers are ideal for data centers, private servers, and edge deployments.

The servers have multiple computing nodes, featuring either energy-efficient processors (Rockchip RK3588, RK3576, or SOPHON BM1688) or high-performance NVIDIA Jetson modules (Orin Nano, Orin NX). With 60 to 1000 TOPS AI power, the CSB1-N10 servers can handle the demands of large AI models, including language models like Gemma-2B and Llama3, as well as visual models like EfficientVIT and Stable Diffusion.

CSB1-N10 series AI cluster servers

CSB1-N10 series specifications

Features
CSB1-N10S1688
CSB1-N10R3588
CSB1-N10R3576
CSB1-N10NOrinNano
CSB1-N10NOrinNX
SoC
Octa-core BM1688
Octa-core RK3588
Octa-core RK3576
Hexa-core Jetson Orin Nano
Octa-core Jetson Orin NX
Frequency
1.6 GHz
2.4 GHz
2.2 GHz
1.5 GHz
2.0 GHz
Number of Nodes
10 nodes + 1 control node
Control Node Processor
RK3588, 2.4 GHz
AI Computing Power (INT8)
160 TOPS
60 TOPS
400 TOPS
1000 TOPS
RAM per Compute Node
8GB LPDDR4 (optional up to 16GB)
16GB LPDDR4 (optional up to 32GB)
8GB LPDDR4 (optional up to 16GB)
8GB LPDDR5
16GB LPDDR5
Storage per Node
32GB eMMC (optional up to 256GB)
256GB eMMC (optional 16GB and up)
64GB eMMC (optional up to 256GB)
256GB PCIe NVMe SSD
Video Encoding/Decoding
160 channels of H.265/H.264 1080p@30fps decoding, 100 channels encoding
10 channels of H.265 8K@30fps decoding, 10 channels encoding
Storage Expansion
SATA 3.0/SSD hard drive slot
Ethernet
2 × 10G Ethernet (SFP+)

2 × Gigabit Ethernet (RJ45)

1 × Gigabit Ethernet (RJ45, MGNT is used as BMC management network)
Console
RJ45 console port
Display
VGA up to 1080p60
USB
2 × USB3.0
Button
Reset, UID, power button
Misc
1 × RS232 (DB9, baud rate 115200)
1 × RS485 (DB9, baud rate 115200
Cooling
6 high-speed fans
Power Supply
550W (non-hot-swappable)
Form Factor
1U rack-mounted
Physical Dimensions
420×421.3×44.4mm
Operating Temperature
0°C ~ 45°C
Deep Learning Frameworks
TensorFlow, PyTorch, PaddlePaddle, ONNX, Caffe
cuDNN frameworks: TensorFlow, PyTorch, MATLAB, etc.
Large Model Support
Transformer-based LLMs
LLaMa3, Phi-3 Mini, and vision models
CSB1 N10 series AI cluster servers interfaces
CSB1-N10 series AI cluster server interfaces

All CSB1-N10 AI servers have the same interfaces, and the only differences are the CPU, memory, storage, multimedia, AI capabilities, and related software support. So it’s likely Firefly has made Rockchip system-on-modules compatible with NVIDIA Jetson SO-DIMM form factor, and indeed we previously noted that Firefly designed Core-1688JD4, Core-3576JD4, or Core-3588JD4 CPU modules with BM1688, RK3576, and RK3588 SoCs, so those are likely the ones used in the CSB1-N10 servers.

In terms of software, the CSB1-N10 series AI cluster servers feature a BMC system with Redfish, VNC, NTP, and monitoring for real-time management. Additionally, they have support for deep learning frameworks like TensorFlow, PyTorch, PaddlePaddle, ONNX, and Caffe, with cuDNN acceleration for certain models. They also support Docker containers that enable private deployment of large models like LLaMa3, Phi-3 Mini, EfficientVIT, and Stable Diffusion. Note the wiki entries for the CSB1-N10 series are all empty, so you may want to refer to the documentation for the respective CPU modules mentioned above.

Previously we have written about various clusters and cluster motherboards including the Lichee Cluster 4A mini-ITX, the Mixtile Cluster Box, and the Turing Pi 2.5 mini-ITX motherboard, but the solutions provided by Firefly are more powerful and can be easily integrated into a rack.

The price ranges from $2,059.00 for the CSB1-N10R3576 to $14,709.00 for the high-performance CSB1-N10NOrinNX Computing Server with 1000 TOPS of processing power. Purchasing links for all models are available on the products page, and you’ll also find some additional information on the datasheets for the Rockchip/SOPHGO and NVIDIA cluster servers.

Share this:

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress

Radxa Orion O6 Armv9 mini-ITX motherboard
Subscribe
Notify of
guest
The comment form collects your name, email and content to allow us keep track of the comments placed on the website. Please read and accept our website Terms and Privacy Policy to post a comment.
0 Comments
oldest
newest
Boardcon EM3562 Rockchip RK3562 SBC with 8 analog camera inputs