Running large AI models or handling many simultaneous requests can overwhelm a typical server. When performance drops, projects stall and costs rise because you need more hardware or cloud time. The Firefly CSC2-N48SPK3 gives you a compact, powerful rackmount server that handles heavy AI workloads without the need for multiple boxes. It brings high-speed RISC-V processing, built-in management tools, and reliable power so you can keep your services fast and your budget under control.
Firefly CSC2-N48SPK3, K3 x 48(16GB+128GB)+BMC3588 8G RISC-V 2U Rackmount Server with Firefly
Meet the server that turns a crowded data centre into a single, efficient unit. It packs 48 AI-focused compute nodes, each with an 8-core K3 RISC-V processor, plus a control node powered by an RK3588. The machine is built for AI inference, multimodal algorithms, and large language models up to 30 B parameters. If you are tired of latency spikes and complex scaling, this Firefly server delivers steady throughput and easy remote management.
Key Features
- 2U rackmount form factor fits standard data-centre shelves.
- 48 distributed compute nodes each with 8-core 2.4 GHz RISC-V K3 processor.
- Control node with octa-core RK3588 delivering 6 TOPS AI performance.
- 32 GB LPDDR5 memory per node (total 1.5 TB) for large model loading.
- 128 GB UFS 2.2 storage per node plus optional M.2 NVMe SSD expansion.
- Integrated web-based BMC with Redfish and CLI support.
- 12 high-speed cooling fans for consistent thermal management.
- Two hot-swappable redundant AC power supplies.
- HDMI output with 1920 x 1080 max resolution for local console.
- Two USB 3.0 ports, one supporting OTG and BMC firmware upgrades.
Specifications
| AI Performance |
2880 TOPS (60T × 48, SquareINT4), Supports RVV Vector Instruction Set,
Native FP8 Inference, Multimodal Algorithm Acceleration, Single Node Supports 30B Local Models
with Performance Exceeding 10 Tokens/S@30B, Supports Deployment of All AI Algorithms and Models
|
| Server Form |
2U Rackmount Computing Power Server |
| Number of Nodes |
48 Distributed Computing Nodes + 1 Control Node |
| Compute Nodes |
8-core X100 64-bit RISC-V AI Processor (SpacemiT Key Stone K3),
Maximum Frequency: 2.4 GHz, 130K DMIPS per Node, RVA23 Profile Compliant,
Single-Core SPECint2006 > 9.0/GHz, ARM Cortex-A76 Equivalent Performance
|
| Control Node |
Octa-core 64-bit RK3588 Processor, Maximum Frequency: 2.4 GHz, AI Performance: 6 TOPS |
| Fan Module |
12 High-Speed Cooling Fans |
| Installation Requirements |
IEC 297 Universal Cabinet Installation (19-inch Width, 800 mm Depth or Above),
Retractable Slideway Installation, Cabinet Front-to-Rear Hole Distance: 543.5 mm–848.5 mm
|
| BMC |
Integrated Web-Based BMC Management System, Supports Monitoring, Configuration,
Alarming, Remote Operation & Maintenance, Virtual Replacement Management,
CLI Command Line, Redfish, Secondary Development Tools
|
| Console |
1 × Console (RJ45, BMC Debug Serial Port, Baud Rate: 115200) |
| Memory |
32GB LPDDR5 × 48, Optional Capacities: 8GB / 16GB / 32GB per Node |
| Storage |
128GB UFS 2.2 × 48 |
| Storage Expansion |
48 × M.2 2280 PCIe NVMe SSD (Optional) |
| Codecs |
Video Encoding: 4K@60fps H.264, Video Decoding: 4K@120fps H.264 / VP9 |
| Power |
2 × AC Redundant Power Supplies, Hot-Swappable |
| Display |
1 × HDMI, Maximum Resolution: 1920 × 1080 |
| Dimensions |
724.0 × 430.0 × 88.8 mm (L × W × H), Standard 2U Rack Server |
| Weight |
Net Weight: 23.1 kg, Gross Weight (with Packaging): 25.3 kg |
| Environment |
Operating Temperature: 0°C to 35°C, Storage Temperature: -40°C to 60°C,
Operating Humidity: 5% to 80% RH (Non-Condensing)
|
| USB |
2 × USB 3.0, Lower USB Supports USB 3.0 OTG and BMC Upgrade via USB Flash Drive |
| Buttons |
1 × Reset Button, 1 × Power Button, 1 × Restart BMC Button |
