Skip to main content

Alibaba T-Head Zhenwu PPU

Vendor: Alibaba

Category: GPU Graphics Processor

Architecture: T-Head RISC-V

Introduction​

Alibaba's T-Head Semiconductor subsidiary. Zhenwu 810E PPU (dedicated processor) uses 5nm process, performance comparable to NVIDIA H20. Officially launched in January 2026, cumulative shipments of 470,000 units serving 400+ customers, covering Alibaba's Qwen series model training and inference. Hanguang 800 AI inference chip is an earlier product. Approximately 265,000 units shipped in 2025, domestic second tier.

Specifications​

ModelComputeMemoryInterfaceTDPProcess
C920RISC-V AI extensionsExternal DDRSoC integration5-15W7nm

Official Website​

Visit Official Website

Driver Downloads​

Linux​

OS Support​

WindowsLinuxmacOSAndroid
❌✅❌❌

Version History​

VersionRelease DateDescription
T-Head SDK 1.02024AI inference acceleration library

Performance Benchmarks​

ModelTaskPerformance Metric
C920Edge InferenceRISC-V AI extension acceleration
C920Speech/NLP InferenceSuitable for endpoint scenarios

Pricing​

ModelReference PriceNotes
C920Provided with SoCT-Head RISC-V chip

Quick Installation​

Linux​

# T-Head C920 uses RISC-V architecture
# AI extensions programmed via THead SDK
# Obtain SDK from T-Head Developer Community

Code Examples​

AI development for T-Head Xuantie C920 is primarily through the THead AI SDK. The ecosystem is currently new with limited examples.

Architecture Highlights​

  • RISC-V + AI Extensions: T-Head proprietary RISC-V processor with added AI vector extension instructions
  • Endpoint Positioning: Primarily for IoT and edge device AI inference
  • Open-Source Ecosystem: Leveraging RISC-V open-source ecosystem, THead provides a complete toolchain

Model Compatibility​

Model/FrameworkSupportNotes
TFLite Micro⚠️RISC-V porting in progress
Endpoint Models⚠️Ecosystem developing
Speech/NLP⚠️Alibaba scenario-optimized

If you're evaluating alternatives, the following products may also fit your scenario: