CSA1-N8 – 1U AI Server with Eight BM1684X Compute Nodes

1U AI Server with Eight BM1684X Compute Nodes

Maker, Prototyping, Open-Hardware & Developer Boards

Manufacturer part number
CSA1-N8
Manufacturer
Firefly

Firefly CSA1-N8 1U rack server with eight BM1684X compute nodes and an RK3588S control node, designed for AI inference and video analytics.

Product overview

Overview. The Firefly CSA1-N8 is a 1U rack-mounted AI server integrating eight distributed BM1684X compute nodes and one RK3588S control node, for AI inference and video analytics. Each compute node is an octa-core 64-bit processor running up to 2.3 GHz, delivering 32 TOPS INT8 per node, or 256 TOPS INT8 in full configuration.

Main characteristics. 16 GB LPDDR4 memory per node (8 × 16 GB, 12 GB configuration optional). 64 GB eMMC storage per node (8 × 64 GB, 32 GB or 128 GB optional). One 3.5-inch or 2.5-inch SATA 3.0 drive bay with hot-swap support. Six high-speed cooling fans.

Interfaces and connectivity. Two 10G SFP+ Ethernet ports and two Gigabit Ethernet RJ45 ports, one of them a management network port. Two USB 3.0 HOST ports, one Type-C USB 3.0 OTG port for processor core board debugging, one HDMI 2.0 output with a maximum resolution of 1080P, one SIM card slot and two 4G antennas.

Power and environment. Normal power consumption of 300 W, maximum 430 W. Operating temperature 0 °C to 42 °C, storage temperature -40 °C to 60 °C, operating humidity 20% to 80% RH non-condensing. Server net weight 6.7 kg, 8.9 kg with packaging.

Typical uses. Private deployment of large language models (Gemma, Llama, Qwen, Deepseek-R1 and others), vision large models (ViT, SAM, Grounding DINO) and image generation models (Flux, Stable Diffusion, Stable Diffusion XL). Video stream processing with H.264 decoding of 8×4K@25fps or 32×1080P@25fps. Data management, backup and video surveillance through the storage bay.

Integration. Installation in an IEC 297 universal 19-inch cabinet 800 mm deep or more, or on retractable slideways with a front-to-rear mounting hole distance of 543.5 mm to 848.5 mm. BMC management system with web interface, supporting Redfish, VNC and NTP, and redevelopable. Compatible with TensorFlow, PyTorch, PaddlePaddle, ONNX and Caffe frameworks, with Docker container management.

See also

Technical questions

Ask a question

No published question about this part yet. Questions are reviewed before publication.

Request a quote