Huawei Accelerates AI Development with Ascend 910C Chip | Nvidia hardware | Hardware definition and examples | Gpu benchmark | Turtles AI

Huawei is reducing dependence on Nvidia technology with the Ascend 910C, an AI processor that achieves 60 percent of the performance of Nvidia’s H100 in inference tasks. Although AI training remains a weak point, Chinese technological progress is evident.

Key points:

• Inference performance: Ascend 910C offers 60% of the power of the H100.
• Challenge in training: Nvidia maintains the lead with its advanced ecosystem.
• Software optimizations: DeepSeek facilitates integration of Huawei chips with PyTorch.
• Technology independence: China continues to develop alternatives to reduce dependence on the West.

Huawei is continuing its strategy of enfranchising itself from U.S. technology with the Ascend 910C, an AI chip developed by HiSilicon that offers an inference level equal to 60 percent of Nvidia’s H100, according to tests conducted by DeepSeek. Although the processor is not the most suitable for training advanced AI models, it still demonstrates a remarkable capability in inference, a crucial area for the application of AI in practical fields. DeepSeek’s native support for Ascend processors and optimization of CUNN code enable the 910C’s capabilities to be fully exploited, simplifying the transition from Nvidia chips with a smooth conversion from CUDA to CUNN. China has made significant strides in the semiconductor arena, despite U.S. restrictions and technological limitations in advanced manufacturing. The Ascend 910C uses a chiplet architecture with 53 billion transistors and, unlike its predecessor produced with TSMC’s N7+ technology, employs a 7nm SMIC manufacturing process called N+2. Although Nvidia continues to dominate the AI training market with its established hardware-software ecosystem, the rapid development of the Chinese industry could change the balance in the coming years. DeepSeek R1, one of the newest Chinese AI models, has excelled in various reasoning tasks and, although trained on Nvidia H100, uses Ascend 910C for inference, with surprising results. Predictions indicate that as AI architectures evolve, the weight of the Nvidia ecosystem may shrink, favoring alternatives such as Huawei’s. Chinese progress is not limited to software: Huawei’s upcoming AI chip, expected to coincide with Nvidia’s GTC, could represent a breakthrough in the industry. Meanwhile, Intel is going through a difficult period, posting losses and shrinking profit margin, while China consolidates its position in AI hardware.

Global competition in AI is fiercer than ever, with China determined to close the technology gap and come up with increasingly competitive solutions.