Chinese GPU Manufacturers Enhance Support for DeepSeek’s AI Models on Local Systems, Accelerating the AI Competition

Chinese GPU manufacturers are making significant strides in the global AI landscape by enabling support for running DeepSeek’s R1 large language model (LLM) on local systems. This move signals their ambition to compete in the fast-evolving AI sector.

Advancements in Chinese GPUs: Moore Threads and Baidu’s Kunlun Champion DeepSeek’s AI Models

Recently, China has been at the forefront of the AI industry, not only with the introduction of models that rival those from established players like OpenAI but also through the enhancement of AI hardware capabilities. Noteworthy developments include Huawei’s preparation to take on NVIDIA’s stronghold with its Ascend 910C AI chip, alongside notable contributions from Chinese GPU makers.

Among the leaders in this initiative are Moore Threads and Baidu’s Kunlun, both of which have successfully enabled support for DeepSeek’s R1 model using their consumer GPUs. This advancement is crucial as it propels the race for enhanced computational power in AI.

Moore Threads: Enhancing Local AI Processing

Moore Threads has introduced support for DeepSeek’s distillation models via a deployment service. Their MTT S80 and MTT S4000 GPUs are now compatible with these models, with the MTT S4000 specifically engineered for handling demanding workstation workloads. Furthermore, the company has made these models accessible on its in-house KUAE cluster, designed meticulously for AI tasks and featuring MTT S4000 GPUs.

This breakthrough by Moore Threads is anticipated to encourage professional users to adopt their GPUs, given the robust capability to run DeepSeek’s AI models locally. Additionally, the MTT S80 and MTT S4000 can be utilized for inference workloads, showcasing the manufacturer’s extensive support. However, the performance metrics of DeepSeek’s models on these GPUs remain unverified, but it is unlikely they will match the output of AMD and NVIDIA’s offerings.

Baidu's AI Cluster — Image Credits: MyDrivers

Baidu’s Initiative: The Kunlun Core P800 AI Chip

Baidu has made significant headway with its in-house AI cluster equipped with Kunlun Core P800 AI chips. Reports from MyDrivers suggest that the Core P800 outperforms similar mainstream GPUs by 20-50%, supports 8-bit inference, and is designed for low deployment and maintenance costs. This AI chip is fully compatible with DeepSeek’s V3/R1 AI models, making deployment straightforward.

The scale of Baidu’s AI cluster is impressive, comprising 30, 000 Core P800 AI GPUs, with plans for full operational capability soon. These advancements illustrate how Chinese GPU manufacturers are not only overcoming international obstacles but also reinforcing their commitment to developing a sustainable and robust domestic hardware infrastructure.

Source & Images