Overview
CSA1-N8S1684X AI Computing Power 1U Server is an AI inference server from Firefly built on BM1684X, rated by Firefly at 256 TOPS (INT8) peak computing pow. It carries 16GB LPDDR4 × 8 (Number of Compute Nodes). Firefly publishes a fixed price for this configuration and sells direct.
Key Features
- AI compute: 256 TOPS (INT8) peak computing pow
- Accelerator: BM1684X
- Memory: 16GB LPDDR4 × 8 (Number of Compute Nodes)
- Storage: 32GB eMMC × 8 (Number of Compute Nodes), 3.5-inch SATA3.0 hard drive × 1
- OS: Linux
Ideal For
teams running on-premises inference for privately deployed models without buying datacenter GPUs
Number of nodes
8 compute nodes (up to 64 ARM cores) + 1 control node
Ai computing power
256 TOPS (INT8) peak computing power, 128TFLOPS (FP16/BF16) computing power,16TFLOPS (FP32) high-precision computing power
Bmc
Integrated BMC management system, providing a web-based management interface, BMC management system can be redeveloped
Deep learning
TensorFlow/PyTorch/PaddlePaddle/Caffe/ONNX/MXNet/DarkNet
Cpu
Compute node: BM1684X octa-core (A53) 64-bit processor, with a maximum frequency of 2.3 GHz.Control node: RK3588S octa-core (4×Cortex-A76+4×Cortex-A55) 64-bit processor, with a maximum frequency of 2.4GHz.
Ram
16GB LPDDR4 × 8 (Number of Compute Nodes)
Storage
32GB eMMC × 8 (Number of Compute Nodes), 3.5-inch SATA3.0 hard drive × 1
Power
600W AC power supply (Input: 100V AC ~ 240V AC, 50Hz ~ 60Hz, 4A ~ 8A)
Power consumption
Normal: 190W ~ 220W, Max: 370W ~ 450W
Display
HDMI2.0 × 1 (1080P, main processor core board display)
Os
Linux
Internet
10G Ethernet (SFP+) × 2, Gigabit Ethernet × 2 (RJ45, 1 control node port, 1 common network port), 4G LTE/5G network (optional)
Usb
USB3.0 HOST × 2, Type-C × 1 (For processor core board debugging)
Area
Shipping time
Europe
3 - 7 business days
Asia, Oceania
3 - 7 business days
North America
3 - 7 business days
South America
5 - 10 business days
Brand
Firefly
Prices may vary. Verify on vendor site.
Quick Specs
- Number of nodes
- 8 compute nodes (up to 64 ARM cores) + 1 control node
- Ai computing power
- 256 TOPS (INT8) peak computing power, 128TFLOPS (FP16/BF16) computing power,16TFLOPS (FP32) high-precision computing power
- Bmc
- Integrated BMC management system, providing a web-based management interface, BMC management system can be redeveloped
- Deep learning
- TensorFlow/PyTorch/PaddlePaddle/Caffe/ONNX/MXNet/DarkNet
- Cpu
- Compute node: BM1684X octa-core (A53) 64-bit processor, with a maximum frequency of 2.3 GHz.Control node: RK3588S octa-core (4×Cortex-A76+4×Cortex-A55) 64-bit processor, with a maximum frequency of 2.4GHz.
- Ram
- 16GB LPDDR4 × 8 (Number of Compute Nodes)