Iluvatar CoreX demonstrated that its Tiangai 300 graphics processing unit (GPU) can equal or exceed the performance of Nvidia’s H200 in real-world artificial intelligence tasks.
During a live demonstration for mainland Chinese media outlets on 28 October, the company showed the Tiangai 300 outperforming the H200 in complex multimodal operations. In a text-to-image test, the Tiangai 300 achieved a 100 per cent completion rate with a throughput of 0.101 images per second, according to Xinzhixun.
The H200 recorded a success rate of 87.5 per cent and a speed of 0.086 images per second during the same evaluation. Xinzhixun also noted that the Tiangai 300 required 0.64 seconds more to finish the text-to-image task than the H200. Wallstreetcn reported that the Tiangai 300 beat the H200 in mixture-of-experts (MoE) throughput by 15 per cent.
The same outlet stated the Tiangai 300 reduced first-token latency by up to 30 per cent when processing long-text. Nvidia began shipping its first H200 data-centre processors to China in August, following the initial launch of the chip in late 2023.
Iluvatar unveiled a multi-year architecture road map in January with the goal of surpassing Nvidia’s advanced Rubin platform within two years. The company stated that the Tiangai series has already reached deployments on a 10,000-card scale. Iluvatar went public in Hong Kong earlier this year.
Tongyang TY1200 processor outperformed Nvidia’s Thor T5000 by 20 per cent in specific large language model tests, according to Wallstreetcn.
