Alibaba has successfully integrated its XuanTie C950 RISC-V chip, manufactured by TSMC using a 5nm process, with its Qwen-3.8 27B large language model. This integration allows the model to run natively on the chip, achieving a performance of 30 tokens per second. This development signifies a significant step towards vertical integration for Alibaba in the AI hardware and software space, potentially reducing reliance on traditional GPU architectures. AI
IMPACT Demonstrates potential for RISC-V architecture in running LLMs, reducing reliance on GPUs and enabling greater hardware-software integration.
RANK_REASON Research milestone involving a novel chip architecture running a large language model natively.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →