Huawei and DeepSeek release open source AI tools in a bid to lower Nvidia exposure — but will programmers make the switch?
DeepSeek's two new repositories already have hundreds of GitHub stars between them.
Huawei and DeepSeek have released open source AI tools in an attempt to reduce their reliance on Nvidia hardware. The companies published two new repositories, DeepGEMM-Ascend for matrix multiplication and DeepEP-Ascend for all-to-all communication, which are crucial for mixture-of-experts models. These are not updates to existing tools but genuinely new projects.
While DeepSeek also updated existing projects like TileKernels, DeepSelect, and FlashMLA for Ascend accelerators, the two new repositories received fewer stars on GitHub than Nvidia's popular DeepGEMM project. The tools require specific Ascend hardware and specific connectivity, making them limited in use cases. Despite this, the open source offerings aim to ease the switch from Nvidia hardware, as Nvidia's chips are unlikely to be freely available in China due to US export restrictions.
The reputational issue for Huawei around its Ascend chips and software is an ongoing concern, with reports of instability and a steep learning curve.
Written by urgent.news from TechRadar's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
