DeepSeek Open-Sources Six Tools to Boost Huawei Chips Over Nvidia in AI

DeepSeek, an AI model developer based in Hangzhou, has introduced six software modules aimed at establishing a "self-sufficient and manageable" software ecosystem for graphics processing units (GPUs), as stated in a post on their official WeChat account. These tools are similar to previous open-source offerings for Nvidia's AI chips. Notably, one of the releases includes a version of TileLang compatible with Ascend, a specialized programming language that simplifies the creation of high-performance kernels for GPUs and central processing units. Although TileLang primarily supports Nvidia, it now officially accommodates Huawei's Ascend 950 accelerators, featuring "native code generation, automatic scheduling, and synchronization," according to an update on GitHub. This initiative seeks to lessen dependence on Nvidia and strengthen Huawei’s presence in the AI chip sector.