Skip to content

Event

Iluvatar CoreX opens live demonstration of TianGai 300, claims per-card throughput reaches 1.05x H200

1 reports · 1 sources

Iluvatar CoreX held its first live demonstration of the TianGai 300 at a media open day on October 28. According to the company, third-party measured results converted to per-card overall throughput reach 1.05 times that of Nvidia's H200. Information released at the same time also includes: the IXL144 supernode integrates 144 GPUs; the AFD disaggregation scheme for GPU+XPU heterogeneous deployment reduces inference latency by 60%; the edge-side product TongYang TY1200's measured performance surpasses Thor T5000, and it is the first domestic edge-side product to implement GPU-accelerated physical simulation based on MuJoCo-Warp. All the above performance figures come from Iluvatar CoreX.

Report timeline

  1. Iluvatar CoreX opens live demo of Tiangai 300, single-card throughput reaches 1.05x H200

    At its October 28 media open day, Iluvatar CoreX opened the first live demo of the Tiangai 300; third-party testing converted to a single-card overall throughput of 1.05x H200. The concurrently released IXL144 supernode integrates 144 GPUs, and a GPU+XPU heterogeneous deployment AFD disaggregation scheme reduces inference latency by 60%. The edge product Tongyang TY1200 outperforms Thor T5000 in measured performance and is the first domestic edge product to implement GPU-accelerated physics simulation based on MuJoCo-Warp.

    36Kr Embodied Intelligence ·