Huawei challenges Nvidia: the cluster cloudmatrix exceeds expectations | Hardware near me | Hardware definition and examples | 5 basic components of cpu | Turtles AI

Huawei recently presented the Cloudmatrix 384 artificial intelligence cluster, an infrastructure that, according to what was declared, exceeds the GB200 NVL72 system in Nvidia in performance, thanks to the use of 384 Ascode 910C chips. This move highlights the growing competitiveness of the Chinese company in the AI ​​sector, despite the restrictions imposed by the United States.

Key points:

• The Huawei Cloudmatrix 384 cluster integrates 384 ASCEnd 910C chips, offering performance higher than the NVIDIA GB200 NVL72 system.
• The ASCEnd 910C chip reaches 60% of the inference performance of the NVIDIA H100, according to tests conducted by Deepseek.
• Huawei has improved the production of Aschend 910C, aiming for a 60% yield by 2025.
• Huawei’s development of the cann software aims to reduce dependence on the Cuda ecosystem of Nvidia.

The CEO of Nvidia, Jensen Huang, recognized Huawei’s progress in the field of AI, underlining how the Chinese company is quickly filling the technological gap. The Cloudmatrix 384 cluster, which uses 384 Ascend 910C chips, was designed to compete directly with the NVIDIA GB200 NVL72 system, offering performance of 300 petoflops in BF16 calculation, exceeding the 180 petoflops of the NVIDIA system.

The ASCEnd 910C chip, produced with 7 Nm technology from SMIC, represents a significant evolution compared to previous models. According to tests conducted by Deepseek, the Ascend 910C reaches 60% of the NVIDIA H100 inference performance, making it a valid alternative for Chinese companies that face restrictions in accessing Nvidia chips.

Huawei has also improved the production of Aschend 910C, increasing the surrender from 20% to 40%, with the aim of reaching 60% by 2025. This increase in production aims to meet the growing internal demand and to consolidate the Huawei position in the Chinese AI chip market.

To reduce dependence on Nvidia software ecosystem, Huawei has developed its cann framework, compatible with MindSpore, Tensorflow and Pytorch. This strategy aims to facilitate the transition of Chinese companies towards completely national hardware and software solutions, strengthening the country’s technological autonomy.

Despite the challenges imposed by international restrictions, Huawei continues to invest in the development of advanced technologies, trying to offer competitive solutions both in terms of hardware and software. The growing adoption of the Ascend 910C Chips and the Cloudmatrix 384 cluster highlights the company’s commitment to consolidate its position in the AI ​​sector.

The rapid evolution of Huawei’s abilities in the field of AI underlines the importance of carefully monitoring future developments in this strategic sector.