-+ 0.00%
-+ 0.00%
-+ 0.00%
The 2026 Sustainable Global Leaders Conference will be held from October 20 to 22 at the Huangpu World Expo Park in Shanghai. Yang Huiqian, general manager of the intelligent computing business of Shangtang University, attended and delivered a speech. Yang Huiqian pointed out that the speed of performance improvement at the hardware level of domestic chips is remarkable. Judging from the comparison of MFU indicators, some of the current performance indicators of mainstream domestic chips can already be compared to similar Nvidia chips, and the hardware performance alone already has good basic strength. However, the implementation value of computing power still needs to be judged based on the needs of AI scenarios. Domestic chips for inference scenarios have reached the level of being usable and easy to use. Based on Shangtang's implementation practice, high-end flagship chips and domestic mainstream chips are divided into tasks through a PD separation architecture. The economic benefits of this heterogeneous solution are superior to the computing power clusters of the same configuration. Relying on this architecture, the average daily token output of Shangtang's inference business can reach 6 trillion dollars, which is expected to reach 10 trillion tokens in the fourth quarter.
Share
Listen to the news
The 2026 Sustainable Global Leaders Conference will be held from October 20 to 22 at the Huangpu World Expo Park in Shanghai. Yang Huiqian, general manager of the intelligent computing business of Shangtang University, attended and delivered a speech. Yang Huiqian pointed out that the speed of performance improvement at the hardware level of domestic chips is remarkable. Judging from the comparison of MFU indicators, some of the current performance indicators of mainstream domestic chips can already be compared to similar Nvidia chips, and the hardware performance alone already has good basic strength. However, the implementation value of computing power still needs to be judged based on the needs of AI scenarios. Domestic chips for inference scenarios have reached the level of being usable and easy to use. Based on Shangtang's implementation practice, high-end flagship chips and domestic mainstream chips are divided into tasks through a PD separation architecture. The economic benefits of this heterogeneous solution are superior to the computing power clusters of the same configuration. Relying on this architecture, the average daily token output of Shangtang's inference business can reach 6 trillion dollars, which is expected to reach 10 trillion tokens in the fourth quarter.
Disclaimer:Webull uses external vendor Google Translation Service for news translations where we endeavour to ensure these are correct, however, we recommend that you please double-check this information accordingly. Webull is not responsible for translation errors or issues.
What's Trending