A new artificial intelligence (AI) a framework created by groups affiliated with China Tsinghua University is claimed to help decrease dependence on Nvidia Chips designed for AI model inference represent the newest initiative by the nation to boost technological independence.
Chitu, a highly efficient inference framework designed for large language models (LLMs), is capable of functioning on Chinese-made chips. This development poses a challenge to Nvidia’s Hopper series GPUs, which have been dominant in supporting specific models like the aforementioned ones. DeepSeek-R1 , according to a joint statement by start-up Qingcheng.AI and a team led by computer science professor Zhai Jidong at Tsinghua University on Friday.
AI frameworks act as foundational elements for advanced intelligent systems, providing a suite of libraries and tools that allow developers to create, train, and test intricate models effectively.
Are you looking for insights into the most significant issues and developments globally? Find your answers here. SCMP Knowledge Our latest platform features handpicked content including explainers, FAQs, analyses, and infographics, all provided by our acclaimed team.
Since last Friday, the Chitu framework has been released as open-source software. According to the firm, it supports popular models such as those developed by DeepSeek and Meta Platforms’ Llama lineup.
When tested with the full-strength version of DeepSeek-R1 using Nvidia's A800 GPUs, the framework achieved a 315 per cent increase in model inference speed while reducing GPU usage by 50 per cent compared to foreign open-source frameworks, the company said.
This initiative is part of a larger push by Chinese artificial intelligence firms aiming to reduce their reliance on Nvidia, which produces high-performance GPUs affected by U.S. export restrictions. The company is prohibited by Washington from supplying its cutting-edge H100 and H800 processors from the Hopper lineup to customers based in China.
The growth of Hangzhou’s DeepSeek, which has created its AI models with significantly lower costs and fewer computational resources compared to their western counterparts, has sparked discussions regarding a possible decrease in the need for Nvidia GPUs.
Qingcheng.AI was established in 2023 by Zhai along with his students from Tsinghua University, where Zhai took on the role of chief scientist.
Supported by Beijing’s municipal fund dedicated to the artificial intelligence sector, this startup has joined forces with leading Chinese GPU producers such as Moore Threads , Enflame and Iluvatar CoreX, as stated by CEO Tang Xiongshao during an interview with Chinese media outlets last year.
Several other technology firms in China have stepped up their initiatives to decrease dependence on overseas technologies after the surge of support for DeepSeek.
In February, Infinigence AI—a computing infrastructure platform provider backed by experts from Tsinghua University and financial support from leading Chinese technology companies—announced its efforts towards foster collaboration within the nation's top seven AI chipmakers: Biren Technology , Hygon Information Technology, Moore Threads, MetaX, Enflame, Iluvatar CoreX, and Huawei Technologies' Ascend .
In a recently published research paper, experts from ByteDance, which owns TikTok, revealed a substantial 170 percent improvement in the training process for large language models through their enhanced system. This optimized setup has been deployed across certainByteDance operational areas and reportedly leads to savings equivalent to millions of GPU hours, as stated by the firm.
More Articles from SCMP
What insights into Beijing's assimilation efforts can be gleaned from the Chinese Islamic Association's relaunch?
Within the Hong Kong Ballet's latest offering for senior citizens
Bridgerton's newest romantic interest, Yerin Ha, discusses her upbringing in an acting family, split between Australia and South Korea.
China repatriates 3,000 suspects from Myanmar’s cybercrime haven following 'no safe havens' pledge.
The article initially appeared on the South China Morning Post (www.scmp.com), which serves as the premier source for news coverage of China and Asia.
Copyright © 2025. South China Morning Post Publishers Ltd. All rights reserved.
