每个集群都包含几十万张 GPU,并在集群之间配备了巨大的带宽,这是一套由 OpenAI 和微软共同设计的架构。 算力决定了我们能训练出多聪明的模型,以及我们是否能够扩展产能来满足远超供给的需求。 我们非常期待让这座 AI ‘超级工厂’投入使用。 hundreds of thousands of GPUs per cluster, with massive bandwidth between clusters, co-designed between OpenAI and Microsoft. compute determines how smart of a model we can produce and if we can scale to meet our oversubscribed demand, excited to put this AI superfactory to use.

Greg BrockmanPresident & Co-founder, OpenAI

观点原文

每个集群都包含几十万张 GPU,并在集群之间配备了巨大的带宽,这是一套由 OpenAI 和微软共同设计的架构。

算力决定了我们能训练出多聪明的模型,以及我们是否能够扩展产能来满足远超供给的需求。

我们非常期待让这座 AI ‘超级工厂’投入使用。

hundreds of thousands of GPUs per cluster, with massive bandwidth between clusters, co-designed between OpenAI and Microsoft.

compute determines how smart of a model we can produce and if we can scale to meet our oversubscribed demand, excited to put this AI superfactory to use.

OpenAI

相关观点