美国芯片巨头英伟达近日在其官网上线了针对中国市场特别推出的RTX 5880 Ada工作站显卡,该显卡主要面向消费级和AI训练等专业领域。然而,与旗舰级RTX 6000显卡相比,这款显卡的单精度浮点性能下降了约24%,整体性能降低了近1/4,其实际表现接近次旗舰性能的RTX 5000。

据行业人士透露,英伟达的这款消费级RTX 5880和4090D显卡在AI模型训练方面的表现有限,仅能用于70亿-650亿参数的模型,无法训练万亿级参数模型;而在推理方面,4090D显卡的表现则相对稳定。

This news article title: Nvidia Releases “China-Exclusive” Chips: Performance Drops by Nearly 25%
Keywords: Nvidia, China-Exclusive, Chip Performance

News content:
The American chip giant Nvidia recently launched a special version of the RTX 5880 Ada workstation graphics card for the Chinese market on its official website, targeting professional fields such as consumer-grade and AI training. However, compared to the flagship RTX 6000 graphics card, the performance of this graphics card has decreased by about 24% in single-precision floating-point operations, and the overall performance has decreased by nearly one-quarter, with actual performance close to that of the RTX 5000, a lower-end model.

Sources in the industry reveal that the consumer-grade RTX 5880 and 4090D graphics cards from Nvidia have limited performance in AI model training, suitable for models with parameters of 70 billion to 650 billion, but unable to train trillion-parameter models. In terms of inference, the 4090D graphics card performs relatively stably.

【来源】https://www.tmtpost.com/6868819.html

Views: 1

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注