GPU: Chinese Companies Rally Behind DeepSeek

Chinese consumer GPU manufacturers are actively supporting the implementation of the DeepSeek R1 LLM model on local systems.

This is seen as a major step forward in the global AI race. Two prominent companies in this field are Moore Threads and Baidu, with Baidu's Kunlun GPUs now capable of running full DeepSeek AI models on servers.

Moore Threads MTT S4000 GPU Series.

China has made significant strides in the AI ??industry, not only by releasing models that compete with OpenAI's products but also by surpassing them in hardware capabilities. Huawei recently prepared to challenge NVIDIA's dominance with its Ascend 910C AI chip. Now, Moore Threads and Kunlun have added support for DeepSeek's R1 model to boost its computing power.

Moore Threads has announced support for DeepSeek models through its deployment service, which is compatible with the MTT S80 and MTT S4000 consumer GPUs. The MTT S4000 GPUs are specifically designed for workstation workloads. The company has also deployed DeepSeek models on KUAE, an internal cluster dedicated to large AI workloads and using MTT S4000 GPUs. This is a notable achievement, making it easier for professional consumers to adopt Moore Threads products. However, the performance of DeepSeek models on these GPUs is still unclear and may not reach the level of what AMD and NVIDIA offer.

Baidu's Core P800 GPU series.

Baidu, one of China's leading tech companies, has also built an in-house AI cluster with the Kunlun Core P800 AI chip. According to MyDrivers, the Core P800 is 20-50% more efficient than similar mainstream GPUs, supports 8-bit inference, and has low deployment and maintenance costs. The AI ??chip is fully compatible with DeepSeek's V3/R1 AI models, and inference deployment is said to be quite easy. Baidu's AI cluster is expected to include 30,000 Core P800 AI GPUs and will be operational soon.

The achievements of Chinese GPU manufacturers show that they are not affected by global pressure, but instead are focusing on developing domestically produced products, helping them maintain long-term sustainability.