Iluvatar CoreX Tiangai 300 Beats Nvidia H200 in Tests

Iluvatar CoreX Tiangai 300 has been publicly tested as the Chinese manufacturer's AI accelerator. In tests held for press members in Shanghai, the card performed at the level of Nvidia H200 in real artificial intelligence scenarios. Moreover, it exceeded H200 in some measurements. Therefore, the results provide an important reference for the Chinese chip industry looking for an alternative to Nvidia.

First of all, the accelerator, that is, the special processor card that undertakes the heavy calculation load of artificial intelligence models, is the key hardware in this field. Therefore, the efficiency of a card in a real scenario offers a direct comparison.

What Did Iluvatar CoreX Tiangai 300 Achieve in Tests?

First, it is necessary to look at the standard configuration. In this layout, compute and service components were placed separately from each other. Accordingly, the efficiency of a single Iluvatar CoreX Tiangai 300 card reached 1.05 times that of Nvidia H200. Therefore, the Chinese accelerator was 5 percent ahead of its rival in the same measure. The efficiency criterion tells you how much work the card can do in a certain period of time.

It is important to note that the comparison is made per card. Because the figure reflects the difference between a single Tiangai 300 and a single H200. Additionally, measurements were presented through real artificial intelligence working scenarios. Real scenario tests reflect the behavior of the card in daily use, unlike the technical values ​​​​on paper. Therefore, the ratio of 1.05 is an important indicator for the company.

The Difference Widened in Multi-Mode Scenarios

In some more complex scenarios, the Tiangai 300 also outperformed the H200. These are multimodal workloads, that is, tasks where different types of data are processed together. For example, producing images or videos with text commands falls into this group. Therefore, the results show that the card catches up with or even surpasses the H200 in complex tasks.

Large language models were used in these scenarios. Among them was Zhipu AI's GLM-5.3 model. The focus of the test, on the other hand, was the production of images and videos from text prompts. Thus, Iluvatar CoreX Tiangai 300 was also evaluated with heavy workloads.

Why is H200 the Reference Point?

The Nvidia H200 is one of the most powerful boosters allowed to be shipped to mainland China under current US export restrictions. So it provides a natural reference point for comparison. Because the Tiangai 300 was compared to one of the most powerful Nvidia options allowed to be sold to China.

On the other hand, Iluvatar CoreX is just one of the Chinese artificial intelligence chip manufacturers developing alternatives to Nvidia products. For this reason, the Tiangai 300 results also show the current state of the search for alternatives in China.

The most concrete data that stood out at the demonstration in Shanghai was 1.05 times the efficiency of the H200 per single card in the standard configuration. In multi-mode tests, the Iluvatar CoreX Tiangai 300 was ahead of the H200 in image and video production with models such as GLM-5.3. The company has been listed on the Hong Kong stock exchange since January 2026.