The AI wars are heating up with DeepSeek, a Chinese language AI mannequin that claims to surpass US rivals considerably with regards to value effectivity. Its open-source chatbot has propelled the app to the highest place within the App Retailer in 51 international locations, and it is now revealed that it operates on a Huawei AI chip.
The DeepSeek R1 LLM (large-language mannequin) was skilled on Nvidia H100 however makes use of an Ascend 910C chip for inference, which is the motion of utilizing the skilled mannequin to generate responses.
I really feel this ought to be a a lot greater story: DeepSeek has skilled on Nvidia H800 however is operating inference on the brand new dwelling Chinese language chips made by Huawei, the 910C. pic.twitter.com/6IAgQlQ3ou
— Alexander Doria (@Dorialexander) January 28, 2025
The data comes from @Dorialexander, who factors out that Ascend chips are usually not coping with coaching, so the GPU energy necessities are usually not that top.
Nevertheless, the Ascend 910C’s comparatively decrease efficiency limits its suitability for coaching. Huawei plans to deal with this challenge with the upcoming 920C chip goals to compete with Blackwell B200, the main Nvidia chipset for AI operations.
The DeepSeek app on iOS
