SOPHON BM1684X Tensor Computing Processor
Edge tensor processor delivering 32 TOPS INT8 and 16 TFLOPS FP16 with an octa-core Cortex-A53 host and 32-channel HD video decode. Its TPU-MLIR toolchain runs LLaMA, Qwen and ChatGLM locally for on-robot language reasoning.
Inquiry
Chip/SoCBM1684X
AI Compute(TOPS)32 TOPS INT8; 16 TFLOPS FP16/BF16
CPU8x Arm Cortex-A53 @2.3 GHz
Price range (USD)Inquiry
MOQNegotiable
Lead time30-45 days
ExportYes
Full specifications
MANUFACTURER-REPORTED SPECIFICATIONS
| Chip/SoC | BM1684X |
| AI Compute(TOPS) | 32 TOPS INT8; 16 TFLOPS FP16/BF16 |
| CPU | 8x Arm Cortex-A53 @2.3 GHz |
| Memory | LPDDR4X (12 GB on module SKUs) |
| OS | Linux with SOPHON SDK / TPU-MLIR |
| Interfaces | PCIe, GbE, USB 3.0, MIPI CSI |
| Power | Approx. 30 W (typical) |
Supplier
SOPHGO Technologies Ltd. 北京算能科技有限公司
北京, 北京 · est. 2019 · 500-1000
SOPHGO is a Chinese AI computing company founded in 2019 that designs the SOPHON tensor processing family and SG RISC-V CPUs. Its BM1684X delivers 32 TOPS INT8 with 32-channel video decode, powering edge AI boxes, robot perception servers and on-premise large-model inference.
Related models
Sourcing from this guide? Tell us the model, quantity and destination — we'll come back within 24 hours with landed-cost options and honest availability.
Request a Quote— Sourcebotics, sourcing desk