- Published on
DeepSeek and Huawei Collaborate on Developing Ascend Programming Tools and Open-Sourcing Software Infrastructure to R...
- Authors
- Name
- DeepHorus
Huawei announced its next-generation AI processors and supernode computing system two weeks ago, and expects its AI systems to see widespread use in model training next year. DeepSeek confirmed that Huawei is collaborating with the company to develop Ascend programming tools in order to improve a self-reliant and controllable software infrastructure. The two sides are also jointly advancing a supernode solution composed of 128 Ascend 950 chips, with a focus on improving computing and communication capabilities.
In this collaboration, DeepSeek specifically highlighted the role of TileLang. This tool is a high-level, open-source programming language designed for AI chips, with the goal of improving development efficiency and simplifying code logic. DeepSeek stated that the next generation of self-reliant and controllable GPU software ecosystems requires a general-purpose, easy-to-program high-level language, and that TileLang was designed for precisely this purpose.
The collaboration covers both programming infrastructure and the supernode system. Huawei is providing comprehensive support for the relevant development efforts, while DeepSeek is working with Huawei to advance the tools and the solution based on 128 Ascend 950 chips. TileLang has therefore become a key link connecting chip performance with the way developers use the hardware.
In the supernode solution, 128 Ascend 950 chips will form a unified AI computing system. The two sides are currently optimizing the computing and communication capabilities of the solution to better support workloads such as model training. Huawei previously expected its AI systems to achieve widespread application in the field of model training next year.
TileLang uses a high-level programming model designed for AI chips. DeepSeek says that this model is simpler than Nvidia's CUDA and can help reduce development complexity while unlocking hardware performance. The collaboration around TileLang reflects the two sides' efforts to build a self-reliant and controllable GPU software ecosystem.
This collaboration brings Ascend hardware, programming languages, and the supernode system into a single technical roadmap, with the goal of reducing the software ecosystem's dependence on Nvidia's platform. If the relevant tools and solutions continue to advance, Ascend's ease of use and scalability in AI model development and training are expected to improve. Subsequent developments between Huawei and DeepSeek will also become an important indicator for observing the construction of their AI infrastructure ecosystem.
