近日，全球知名芯片巨头英伟达（NVIDIA）悄然在其官网上架了针对中国市场推出的两款特殊规格消费级显卡——RTX 5880 Ada工作站显卡与RTX 4090D。此举旨在应对美国对该公司部分业务实施的禁令影响，满足国内专业用户特别是AI训练和推理领域的特定需求。
新发布的RTX 5880 Ada显卡定位为消费级及AI训推专用设备，相较于旗舰级RTX 6000系列，其单精度浮点性能大幅下降约24%，整体性能损失接近四分之一，预计实际效能将与其次旗舰级别产品RTX 5000相当。然而，有行业内部消息指出，在AI模型训练方面，RTX 5880 Ada仅适用于处理规模在70亿至650亿参数之间的模型，对于需要万亿级参数训练任务则显得力不从心。
相比之下，针对更专业且运算需求高的AI推理场景，英伟达推出了RTX 4090D显卡，这款产品被证实能够稳定支持万亿级参数模型的推理工作，充分体现了其在AI计算领域的强大实力和适应性。此次英伟达为中国市场特别定制的这两款消费级“中国特供版”芯片，无疑将在一定程度上缓解国内相关领域因美国禁令带来的困扰，但同时也显示出业界对高性能AI计算硬件需求的迫切性和多样性。
英语如下：
Headline: “NVIDIA Unveils ‘China-Specific’ Chips: Can Performance Drop Offset the Heaviness of AI Training and Inference?”
Keywords: NVIDIA’s China-specific offering, significant performance drop, restricted AI application
Content: Recently, global chip giant NVIDIA (NVIDIA) quietly launched two special specification consumer-level graphics cards on its official website targeting the Chinese market – the RTX 5880 Ada Workstation GPU and the RTX 4090D. This move is aimed at mitigating the impact of the U.S. sanctions imposed on certain aspects of the company’s business and to cater to specific demands from domestic professional users, particularly in the fields of AI training and inference.
The newly released RTX 5880 Ada is positioned as a consumer-grade and AI-training/inference dedicated device, with single-precision floating-point performance dropping by approximately 24% compared to the flagship RTX 6000 series. This results in an overall performance loss nearing a quarter, indicating that it is expected to be roughly on par with NVIDIA’s second-tier product, the RTX 5000 series. However, inside industry sources point out that the RTX 5880 Ada is only suitable for handling models with parameter sizes ranging from 7 billion to 650 billion parameters; for trillion-parameter training tasks, it appears to fall short.
On the other hand, NVIDIA has introduced the RTX 4090D specifically for more professional AI inferencing scenarios with higher computational requirements. The card has been confirmed to stably support trillion-parameter model inference tasks, demonstrating the strong computing power and adaptability of NVIDIA in the AI domain. The two consumer-level ‘China-specific’ chips tailored for the Chinese market by NVIDIA will undoubtedly alleviate some of the challenges faced by related sectors in China due to the U.S. ban, yet they also underscore the urgency and diversity of high-performance AI compute hardware demands in the industry.
【来源】https://www.tmtpost.com/6868819.html
Views: 1