Tianshu Zhixin Debuts TianGai300 GPU at H200-Level Throughput, 1.05x Single-Card Performance
At its 2026 media open day on October 28, Tianshu Zhixin publicly demonstrated the TianGai300 for the first time, becoming the first domestic general-purpose GPU with measured performance reaching H200 level; third-party tests put single-card overall throughput at 1.05x H200, though test conditions were not yet disclosed. Its IXL144 super-node integrates 144 GPUs as the industry's densest single-cabinet GPU super-node, and a domestic first GPU+XPU heterogeneous deployment AFD disaggregation scheme cut inference latency by 60%. On the software stack, Tianshu Zhixin became the first domestic GPGPU vendor deeply integrated with FlyDSL, and its Tongyang TY1200 outperformed Thor T5000, achieving GPU-accelerated physics simulation on domestic edge devices.
Source: Tianshu Zhixin / Daily AI Brief, Oct 28, 2026 | 2026-10-28
- China Unveils 'StarHub' Plan for Space-Based Computing: 800 Satellites by 2032, Trillion-Yuan Industry
- Action Intelligence Launches Chorus Cross-Embodiment Data Engine, Expanding Robot Training Data 40x
- GTHT Plans 20 Billion Yuan Shanghai Navigation Tech Innovation Fund With AI as Key Focus
- Physical AI Hits Commercial Inflection as Nvidia, Germany and Optimus Bet on a 6 Trillion Dollar Market
- Yiwu's First-Half Export Exceeds 420.7 Billion Yuan for First Time as Christmas Goods Hold 80% of Global Market
- China's Commerce Minister Meets European Parliament Trade Chair, Pledges Stable, Balanced EU-China Ties
- US Section 232 Finds Robot and Industrial Machinery Imports a 'National Security Threat' but Delays Tariffs 180 Days
- Tencent Open-Sources Hunyuan Hy3 With Apache 2.0 License, Free for Commercial Use
- DiffuSphere Open-Sources Clover-35B Diffusion LLM, Claims 5-10x Speed Over Autoregressive Peers
- Wuzhen Summit Open-Source Pavilion Opens as China's Open Models Top Global Call Volume
