The head background
The head background
-
Quad-Engine CAISA Chip
-
High Cost Efficiency
-
3 ms
Nebula Accelerator X3
High Performance, Low Latency and High Cost Efficiency AI Acceleration Solution

-
High Performance
-
Low Latency
-
High Cost Efficiency
-
User-friendly
-
High Versatility
Chip: CAISA
Architecture: CAISA 3.0
Chip Utilization Ratio: Up to 95.4%
Peak performance: 10.9 TOPS
PCIe: PCI Express3.0 x8
Memory: 8GB DDR4
Power Supply: from PCIe slot without AUX power
Cooling: Active Cooling Fan
Temperature: -20℃ ~ 70℃
Dimension: 169.5 mm x 69.6mm
(Standard half-length, half-height, single-slot)
ResNet-50
3.06 ms
1306.93 FPS
ResNet-152
8.68 ms
460.27 FPS
YOLOv3
31.06 ms
125.75 FPS
SSD-ResNet50
21.96 ms
182.16 FPS
U-Net Industrial
74.07 ms
54.01 FPS
Note: Batch=4, INT8 The above CNN models are created by TensorFlow framework
Empowering Industries’AI Upgrading
Smart City
Smart Manufacturing
Smart Grid
Intelligent Safety