Iluvatar CoreX Tiangai 300 GPU Matches NVIDIA H200 Performance
Iluvatar CoreX showcased its Tiangai 300 general-purpose GPU in live operation during a media open day, marking the first time a domestic Chinese GPU achieved measured performance matching NVIDIA's H200. Third-party testing converted the single-card overall throughput to approximately 1.05 times that of the H200. This milestone highlights rapid advancements in China's domestic semiconductor industry as local hardware attempts to narrow the performance gap with market leaders like NVIDIA. Successful deployment of these chips and their supporting architectures could reduce reliance on foreign AI hardware and lower operational costs for domestic cloud and AI providers. Alongside the GPU, Iluvatar CoreX highlighted its IXL144 super-node architecture—which integrates 144 GPUs and 18 CPUs in a single cabinet—and introduced the Tongyang TY1200 edge computing device which features MuJoCo-Warp GPU-accelerated physics simulation. Specific benchmark workloads and testing conditions for the H200 comparison were not publicly disclosed.
## BACKGROUND
General-purpose GPUs (GPGPUs) are specialized processors designed to accelerate heavy parallel computing tasks such as deep learning training and inference. Architecture schemes like attention and mixed-precision operations are critical for handling modern large language models, while edge solutions cater to localized tasks like AI PCs and embodied AI.