Workflow
人工智能研发
icon
Search documents
DeepSeek R2 因芯片问题推迟发布
是说芯语· 2025-08-14 06:28
Core Viewpoint - DeepSeek's launch of its new AI model R2 has been delayed due to issues with Huawei's Ascend chips, highlighting the challenges China faces in achieving technological independence from U.S. technology [3][4][6]. Group 1: Model Development Challenges - DeepSeek has encountered ongoing technical issues while training the R2 model using Huawei's Ascend chips, leading to the decision to use Nvidia chips for training and Huawei chips for inference [4][7]. - The founder of DeepSeek, Liang Wenfeng, has expressed dissatisfaction with the progress of the R2 model and is pushing for increased investment in research and development [8]. - Data annotation for the R2 model has taken longer than expected, contributing to the delay in its release, which is now anticipated within a few weeks [8]. Group 2: Industry Context and Competition - The Chinese government has encouraged tech companies to adopt domestic alternatives to Nvidia products, such as those from Huawei and Cambricon, amid ongoing geopolitical tensions [7]. - Industry experts note that Chinese chips face stability issues, slower inter-chip communication, and inferior software performance compared to Nvidia's offerings [7]. - AI researcher Ritvik Gupta from UC Berkeley commented that models are easily replaceable, with many developers opting for Alibaba's Qwen3 due to its efficiency and flexibility [9]. Group 3: Future Outlook - Despite current challenges, there is optimism that Huawei will eventually adapt to the demands of training AI models with its Ascend chips [10]. - The geopolitical landscape surrounding chip manufacturers like Nvidia remains complex, with Nvidia agreeing to share a portion of its revenue with the U.S. government to resume sales of its H20 chips to China [11].