NVIDIA H800 80GB Pcie 5.0 GPU 高速显卡 计算卡 加速器

NVIDIA H800 80GB Pcie 5.0 GPU 高速显卡 计算卡 加速器

NVIDIA H800 80GB Pcie 5.0 GPU 高速显卡 计算卡 加速器

美国以国家安全问题为由，禁止英伟达向中国客户出售其两款最先进的芯片，即A100和更先进的H100，其背后的目的是减缓中国在半导体和人工智能等关键技术领域的发展。

随后，英伟达发布了A800，同样也是一款阉割芯片，那么目前国内主要使用的就是A800，他与A100的主要差距就是在互联速度上，相差了200GB。

在对大量数据进行人工智能模型训练时，传输速度是相当重要的指标，因此美国要求必须将芯片的传输速率控制在600GB以内。

那么新款H800也是一样的处理方式，将芯片到芯片的数据传输速率降低到旗舰H100速率的一半，而较慢的传输速率也意味着，使用这款芯片就需要花费更多的训练时间。

目前，H800正在被阿里巴巴、百度、腾讯等国内大厂的云计算部门使用。对此，英伟达CEO黄仁勋也表示：“许多年轻的初创公司正在构建大型语言模型，许多公司正投身于生成式人工智能革命，他们可以期待阿里巴巴、腾讯和百度借助英伟达人工智能而拥有出色的云计算能力。”

NVIDIA H800 80GB Pcie 5.0 GPU 高速显卡 计算卡 加速器

The goal behind the US ban on Nvidia selling two of its most advanced chips, the A100 and the more advanced H100, to Chinese customers, citing national security concerns, is to slow China's development in key technology areas such as semiconductors and artificial intelligence.

Subsequently, Nvidia released the A800, which is also a castration chip, so the main domestic use is A800, and the main gap between him and A100 is in the Internet speed, the difference is 200GB.

When training artificial intelligence models on large amounts of data, the transmission speed is a very important indicator, so the United States requires that the transmission rate of the chip must be controlled within 600GB.

The new H800 does the same thing, reducing the chip-to-chip data transfer rate to half that of the flagship H100, and the slower transfer rate means that it takes more training time to use this chip.

At present, the H800 is being used by the cloud computing departments of domestic giants such as Alibaba, Baidu and Tencent. In this regard, Nvidia CEO Jen-Hsun Huang also said: "Many young startups are building large language models, many companies are jumping into the generative AI revolution, and they can expect Alibaba, Tencent and Baidu to have excellent cloud computing capabilities with Nvidia AI."