Deep Learning Acceleration Card SC3

Deep Learning Acceleration Card SC3

One BM1682 integrated, New experience for deep learning accelerating.

top image for mobile

Outstanding Performance by Superior Design

Improved systolic array architecture

Unique NPU scheduling engine for data sharding and data reuse

Flexible and efficient stream computing capacity

Specs

Core Architecture Sophon
Core Frequency 700MHz
Floating point computing speed (FP32) 3 TFLOPS
System Interfaces PCI Express 3.0 x8
Typical/maximum power consumption 50W/90W
Memory 8GB
Cache 16MB
Video decoding 240 FPS H.264 1080p
Image decoding 240 FPS JPEG 1080p

Deep Learning Inference Performance

SC3 Forward
SC3 Throughput

Full stack solution

SC3 software