Cluster there with Huawei Ascend 910c chips can overcome NVL72 with NVIDIA GB200

China seems close to taking another important step towards its independence from external hardware to IA. According to SEMIANALYSIS, the new Huawei clusters, equipped with its Ascend 910C chip, can overcome NVIDIA’s GB200 NVL72.

The Chinese cluster is called Cloudmatrix 384, and the feat was conquered relatively “simply.” Huawei has only placed a large amount of Ascend 910C accelerators in the package – five times more than the amount of the GB200 that Nvidia uses in NVL72. The result can be easily seen in the table below:

The top of the table compares a GB200 with an Ascend 910C, showing a major hardware difference in favor of the Nvidia solution. When comparing clusters, however, Cloudmatrix 384 surpasses NVL72 in all instances.

Obviously, just putting more accelerators in a cluster brings their disadvantages. Scalability and efficiency decrease, while costs go up. SEMIANALYSIS called “terrible” performance numbers by Watt from Cloudmatrix 384 for Nvidia’s solution.

The cost and energy demand part may not be Huawei’s biggest concern, which prepares the Ascend 910C for internal use of the Chinese market and has government support, which increasingly seeks the country’s self -sufficiency.

However, Cloudmatrix 384 should not yet be the final solution to China’s independence in AI technologies for other reasons. Unfortunately for Huawei and its homeland, the company still depends on outside companies for the manufacture of its chips, such as TSMC for example.

Related News:

While China advances in its domestic solutions, those who worry the most should be Nvidia. The company has one of its main markets in the country and does not need to compete only with subsidized local companies, but also facing increasing restrictions from the US to sell there.

Via: WCCFTech