DeepSeek, a Chinese AI firm, has announced a strategic partnership with Huawei Technologies to develop programming tools specifically designed for Huawei Ascend chips. This collaboration is part of a broader movement among Chinese tech companies to reduce their dependence on Nvidia ecosystem, particularly in the face of rising geopolitical tensions and trade restrictions. DeepSeek initiative includes the open-sourcing of programming infrastructure aimed at enhancing the performance of Huawei Ascend platform. This infrastructure will encompass both compute and communication libraries, facilitating improved chip programming efficiency.
The partnership underscores the growing collaboration between Chinese tech entities as they seek viable alternatives to Nvidia dominance in the AI chip market. Huawei recently announced advancements in AI technology, following the unveiling of its new AI processors and supernode computing systems two weeks earlier. The company projects that its AI systems will be widely implemented for model training in the upcoming year. In collaboration with DeepSeek, Huawei has provided comprehensive support to develop a programming infrastructure, resulting in a “supernode” solution that leverages 128 Ascend 950 chips to enhance both computational power and communication efficiency.
Additionally, DeepSeek emphasized the importance of TileLang, an open-source high-level programming language tailored for AI chips, which plays a significant role in boosting development efficiency and streamlining code logic. To create a new generation of independent and self-controlled GPU software ecosystems, establishing a high-level language is essential. This language must be universal, easy to program, and capable of maximizing hardware performance, according to DeepSeek. They introduced TileLang as a solution to this challenge, claiming it provides a simpler programming model compared to Nvidia CUDA.
Leading Chinese Tech Firms Target Nvidia A.I Dominance
In a strategic move towards greater self-reliance in artificial intelligence (AI), leading Chinese tech companies DeepSeek and Huawei have collaborated to develop new software tools specifically designed to enhance the performance of advanced chips. This partnership aims to address a significant challenge that China faces in establishing a domestically-driven AI industry: the need for software that optimizes chip performance. While advanced semiconductors are fundamental for powering AI systems, their efficacy is heavily contingent upon the compatibility and synergy with computer hardware and chip-level software.
As China endeavors to minimize its reliance on American AI technologies, it is striving to close the gap with established leaders in the field, particularly Silicon Valley, where Nvidia has emerged as a dominant force. Nvidia superiority stems not only from its state-of-the-art semiconductor designs but also from the extensive ecosystem of around four million software developers worldwide who utilize its CUDA software platform. DeepSeek has proposed its programming language, TileLang, as a potential alternative to CUDA, asserting that it provides a more user-friendly model that meets the demands of maximizing hardware performance.
TileLang, being open source, offers broader accessibility for modifications and updates, which is essential for fostering an AI ecosystem. Moreover, DeepSeek programming language has been integrated with Huawei AI chips and is viewed as critical in their pursuit of advancing artificial general intelligence technology aimed at surpassing human intelligence across various cognitive tasks. This collaboration is indicative of the broader dynamics at play in China ambitions to rival the U.S. in the AI sector.
Despite significant strides in AI software development, spearheaded by companies like Z.ai and Moonshot AI, there remains a gap in the capabilities of domestic Chinese chip manufacturers. Huawei is actively endeavoring to bridge this disparity. Recently, the company unveiled plans at an event in Shanghai to develop larger, more powerful AI computing systems and to launch new iterations of its Ascend chips by 2028 and 2029. Nonetheless, Huawei has acknowledged its inability to meet the burgeoning domestic demand for AI chips and has opted to prioritize local market needs, indicating a potential restriction on overseas sales.
This decision aligns with the increasing pressure exerted by U.S. export controls, which have curtailed Chinese firms access to Nvidia more advanced chip technologies and essential semiconductor manufacturing equipment. Huawei chairman, Eric Xu, emphasized the urgency of achieving technological self-sufficiency, asserting that China cannot afford a future shaped by external factors, particularly concerning chip supplies. In pursuit of this goal, Huawei unique management strategy involves rotating senior executives through the chairman role every six months, a method aimed at maintaining dynamic leadership and strategic agility.
Summary
- DeepSeek open-sources programming infrastructure for Huawei Ascend platform
- Huawei and DeepSeek jointly advanced supernode solution based on 128 Ascend 950 chips
- DeepSeek ​highlights role of TileLang programming language in project
Pingback: Anthropic $42 Billion Broadcom Deal: How AI Chip Financing Is Changing the Industry - internationalmediawire.com