Reshaping the AI computing landscape: DeepSeek drives the rise of RISC-V
Li Chunqiang, a senior technical expert at Alibaba DAMO Academy, said in an interview that the popularity of the DeepSeek model has brought new development opportunities for the RISC-V architecture. "DeepSeek has significantly reduced the activation parameter ratio through MOE technology, which significantly reduces the computing power required for the model to achieve the same effect. This change provides a new balance point for chip design," said Li Chunqiang. RISC-V is an open source and free CPU instruction set architecture. Compared with traditional instruction set architectures such as x86 and ARM, RISC-V is not only completely free and open, but also supports custom instruction extensions. Therefore, it is regarded as a computing power architecture that is more suitable for the AI era. It is reported that among the 25 standards approved by the RISC-V International Foundation in 2024, more than half are related to high performance or AI.