针对美国禁令，芯片巨头英伟达（NVIDIA）推出了两款面向中国市场消费级的“中国特供版”芯片产品——RTX 5880 Ada工作站显卡。这款显卡主要面向消费级和AI训练推理等专业领域应用。然而，据行业人士透露，这款显卡的单精度浮点性能相比旗舰级RTX 6000降低了24%，整体性能下降接近四分之一，实际表现或接近次旗舰性能的RTX 5000。
对于英伟达消费级RTX 5880和4090D显卡在AI模型训练和推理方面的应用，行业人士表示，RTX 5880仅能用于训练70亿-650亿参数模型，无法训练万亿级参数模型；而4090D显卡则能稳稳运行。这一举措显示出英伟达为了继续服务中国市场，在遵循美国禁令的同时，也在力求保持其产品性能和市场竞争力。
英文翻译：
Title: NVIDIA Releases “China-Exclusive” Chips with nearly 25% Performance Drop
Keywords: NVIDIA, China-Exclusive, Chip Performance
News content:
In response to the U.S. ban, chip giant NVIDIA has launched two consumer-grade “China-exclusive” chip products – the RTX 5880 Ada workstation graphics card. This card is mainly aimed at professional fields such as consumer-grade and AI training and inference. However, industry insiders reveal that the card’s single-precision floating-point performance is 24% lower than that of the flagship RTX 6000, with a nearly 25% drop in overall performance, and its actual performance may be close to that of the mid-range RTX 5000.
As for the application of NVIDIA’s consumer-grade RTX 5880 and 4090D graphics cards in AI model training and inference, industry insiders say that RTX 5880 can only be used for training models with 70 billion to 650 billion parameters and cannot train trillion-level parameter models; the 4090D graphics card can run stably. This move shows that NVIDIA is trying to continue serving the Chinese market while adhering to the U.S. ban and maintaining the performance and market competitiveness of its products.
【来源】https://www.tmtpost.com/6868819.html
Views: 1