Huawei chips are powering DeepSeek AI | Infinium-tech
AI Wars are heating up with Deepsek, a Chinese AI model that claims to cross American competitors when talking about cost efficiency. Its open-source chatbot has topped the app in the app store in 51 countries, and now it is discovered that it operates on a Huawei AI chip.
The Dipsek R1 LLM (large language model) was trained on NVidia H100, but for estimates an ascending uses a 910C chip, which is the action to use trained models to generate reactions.
I think it should be a very big story: Deepsek has trained on Nvidia H800, but 910C is estimated at the new house Chinese chips built by Huawei. pic.twitter.com/6iagqlq3ou
– Alexander Doria (@Dorialexander) 28 January, 2025
Comes from information @DorialexanderThose who say that ascending chips are not working with training, so GPU power requirements are not as high.
However, Ascend limits its suitability for a relatively low performance training of 910C. Huawei plans to address the issue with the upcoming 920C chip, which aims to compete with the major NVIDIA chipset Blackwell B200 for AI operations.
Deepsek app on iOS
Leave a Reply