Abstract:
Chinese artificial intelligence company DeepSeek said on Wednesday that it has cooperated with Huawei to develop programming tools optimized for Ascend chips. DeepSeek announced on its official WeChat account that it will open source the programming infrastructure for Huawei's Ascend platform, including computing libraries and communication libraries. This is the latest sign that Chinese technology companies are deepening cooperation and seeking ecological alternatives to Nvidia.
Huawei announced a new generation of AI processors and super node computing systems two weeks ago, and said it expects its AI systems to be widely used in the field of model training next year. DeepSeek said that Huawei has provided comprehensive support for the development of programming infrastructure, and the two parties have also jointly promoted a "super node" solution based on 128 Ascend 950 chips while optimizing computing and communication capabilities.
DeepSeek specifically mentioned TileLang’s role in the project. TileLang is a high-level open source programming language for AI chips, aiming to improve development efficiency and simplify code logic.
DeepSeek said that to build a new generation of autonomous and controllable GPU software ecosystem, we first need to establish a high-level language that is versatile, easy to program, and can fully utilize the hardware performance; TileLang is designed for this, and its programming model is simpler than NVIDIA CUDA.
Comments