Huawei – Ascend 910C AI Chip Cluster “CloudMatrix” To Outperform NVIDIA’s “Blackwell” GB200 NVL72 Systems; China Catches Up The AI Hardware Gap With The US
Huawei’s AI Cluster With Ascend 910C Chips Will Feature Superior Performance Than NVIDIA’s GB200, But With The Cost of High Power
China’s Huawei is planning to take the AI market once again, this time with its next-gen Ascend AI chip, which is said to disrupt competition from Western powers. In a report by SemiAnalysis, they have managed to dig up details about Huawei’s rack scale architecture and how the Ascend 910C will power China’s most potent CloudMatrix 384 AI cluster. It is said that Huawei’s solution successfully competes with NVIDIA’s most powerful AI server, the GB200 NVL72, showing that China no longer depends on anyone else for their computational needs.
It is claimed that the CloudMatrix 384 (CM384) AI cluster features 384 Ascend 910C chips connected in an “all-to-all topology” configuration. Interestingly, Huawei has managed to cover the architectural flaws by housing five times as many Ascend chips as NVIDIA’s GB200. So, this says Huawei doesn’t care about the costs, performance efficiency, or scalability ratios; instead, the firm simply needs to beat NVIDIA. A CloudMatrix cluster is said to deliver 300 PetaFLOPS of BF16 computing, almost two times higher than GB200 NVL72, and with a higher HBM capacity onboard, the CloudMatrix is said to provide phenomenal results.
Huawei’s rack solution perfectly fits China’s need for a capable domestic computing solution without relying on external factors. However, the only drawback here is the power CloudMatrix 384 is expected to consume, which is said to be 3.9x the power of a GB200 NVL72 and somewhat “awful” perf/watt figures across AI workloads relative to the GB200 AI servers, but this isn’t much of a problem for China at all, since China has the power generation means to light up such AI clusters.
The CM384’s design is intended for Chinese markets alone, and Huawei’s Ascend 910C AI chips will be an ideal fit, given that China doesn’t care much about the expected pricing or power constraints. The real question is how “many” units of the AI clusters Huawei and its partners produce, given that Huawei’s supply chain doesn’t involve Chinese firms alone; it also includes the likes of TSMC, Samsung, and many external partners.
NVIDIA is already having a difficult time sustaining its presence in Chinese markets with the restrictions imposed by the Trump administration. With competition from domestic firms ramping up, Team Green has to find an alternative, and Huawei has managed to come on equal terms with NVIDIA.
READ the latest news shaping the Nvidia market at Newsvidia
Huawei – Ascend 910C AI Chip Cluster “CloudMatrix” To Outperform NVIDIA’s “Blackwell” GB200 NVL72 Systems; China Catches Up The AI Hardware Gap With The US, source
