DeepSeek Open-Sources Huawei Chip Tools as a Simpler Alternative to CUDA
DeepSeek has open-sourced a set of programming tools for Huawei’s Ascend AI chips, including a version of TileLang, a programming language designed as a simpler alternative to Nvidia’s CUDA. This announcement, made on Wednesday in a post on their WeChat account, marks a significant step in the Chinese AI lab's efforts.
The release includes six software modules, offering compute and communication libraries for the Ascend platform, as reported by Reuters. DeepSeek emphasized that Huawei provided full support during development, and the two companies collaborated on a "supernode" system comprising 128 Ascend 950 chips.
"To build a new generation of independent, self-controlled GPU software ecosystems, the first priority is establishing a high-level language that is universal, easy to program, and still capable of reaching the hardware’s full performance potential," DeepSeek stated.
Mirroring Nvidia Tools
The six modules mirror DeepSeek’s existing open-source tools for Nvidia chips, as reported by the South China Morning Post. This move is significant as it expands their support to another major chip manufacturer.
Initially, DeepSeek tested TileLang on older Nvidia chips, as mentioned by The New York Times. Today, they use it as their "core tool" for artificial general intelligence (AGI) development.
Free to Download
The tools are free to download, according to Bloomberg, and the companies plan to collaborate on further projects. DeepSeek also announced plans to deploy at least 160,000 Huawei accelerators in a data center in Inner Mongolia.
Huawei’s Chip Focus
This announcement follows Huawei's introduction of its next-generation Ascend chips, with the company facing AI chip demand exceeding their current production capacity. As reported by The New York Times, Huawei intends to restrict chip sales outside China.
"We cannot accept a destiny where we cannot control our fate," said Eric Xu, Huawei’s rotating chairman.
US export controls further limit Chinese access to Nvidia’s most advanced chips, emphasizing the need for self-reliance in AI technology.