TsingMicro Reconfigurable RPU
Vendor: TsingMicro
Category: ASIC Dedicated Accelerator
Architecture: Reconfigurable Computing (CGRA)
Introduction
TsingMicro, originating from Tsinghua University, focuses on reconfigurable dataflow AI chips (RPU), referred to as the "Transformers of the chip world." Core competitive advantages: high energy efficiency, high flexibility, and low cost. In 2025, it was the only new computing architecture vendor among China's top eight AI accelerator card shippers, with a valuation exceeding 200 billion RMB.
Specifications
| Model | Compute | Memory | Interface | TDP | Process |
|---|---|---|---|---|---|
| TX510 | 10 TOPS (INT8) | External DDR | PCIe / SoC | 3W | 22nm |
| TX210 | 3 TOPS (INT8) | External DDR | SoC integration | 0.5W | 22nm |
Official Website
Driver Downloads
Linux
Related Documentation
OS Support
| Windows | Linux | macOS | Android |
|---|---|---|---|
| ❌ | ✅ | ❌ | ✅ |
Version History
| Version | Release Date | Description |
|---|---|---|
| SDK 2.0 | 2024 | Reconfigurable computing optimization |
Performance Benchmarks
| Model | Task | Performance Metric |
|---|---|---|
| TX510 | Edge Inference | RISC-V CGRA acceleration |
| TX210 | Ultra-low-power Inference | Endpoint scenarios |
Pricing
| Model | Reference Price | Notes |
|---|---|---|
| TX510 | Contact vendor | Chip-level (OEM-oriented) |
| TX210 | Contact vendor | Endpoint solution |
Quick Installation
Linux
# TsingMicro SDK from official source
# Supports RISC-V + CGRA programming
Code Examples
AI development for TsingMicro RPU is primarily through the TSM SDK, supporting CGRA reconfigurable programming. The ecosystem is currently new with limited public documentation.
Architecture Highlights
- CGRA (Coarse-Grained Reconfigurable Architecture): Coarse-grained reconfigurable architecture, flexibly adapting to different AI algorithms
- RISC-V Foundation: Based on RISC-V processor core + CGRA acceleration units
- Endpoint Positioning: Low-power edge AI inference scenarios
Model Compatibility
| Model/Framework | Support | Notes |
|---|---|---|
| Endpoint Models | ⚠️ | TSM SDK support |
| Speech/NLP | ⚠️ | Scenario-optimized |
| LLM | ❌ | Not suitable for large models |
Related Products
If you're evaluating alternatives, the following products may also fit your scenario:
- Hailo-8 / Hailo-15 — Hailo (NPU Neural Processor)
- T-Head Xuantie C920 — Alibaba (GPU Graphics Processor)
- Huawei Ascend — Huawei (NPU Neural Processor)
- Cambricon Siyuan MLU — Cambricon (ASIC Dedicated Accelerator)
- Hygon Shensuan DCU — Hygon (ASIC Dedicated Accelerator)
- Enflame Blaze T20/T21 — Enflame (ASIC Dedicated Accelerator)
- NVIDIA GPU / CUDA — NVIDIA (GPU Graphics Processor)