Meta 近日披露了其最新构建的两个大型GPU集群的详细信息，这些集群将用于训练其先进的Llama3大语言模型。在一篇名为“Building Meta’s GenAI Infrastructure”的博文中，该公司详尽介绍了这两个由24,576个Nvidia Tensor Core H100 GPU组成的集群的硬件、网络、存储、设计、性能和软件配置。这些新集群是基于2022年推出的AI Research SuperCluster (RSC)的扩展和升级。
相较于原先拥有16,000个Nvidia A100 GPU的集群，这一显著的GPU数量增长为Meta的AI研究和开发提供了更强大的计算能力。新的GPU集群设计旨在支持自然语言处理、语音识别和图像生成等领域的尖端AI应用。Meta表示，这种扩展能力使得他们能够处理更大、更复杂的模型，为生成式AI技术的进步奠定了坚实的基础。
这两个全新的GPU集群的推出，不仅展示了Meta在人工智能基础设施领域的持续投入，也预示着未来AI产品开发的潜力将得到显著提升。随着计算能力的增强，Meta有望在语言理解和生成、视觉识别等关键领域实现更多创新，进一步推动人工智能技术的边界。
英语如下：
News Title: “Meta Constructs Massive GPU Arrays: 24k GPUs Power Llama3 Training, Elevating AI Research to New Heights”
Keywords: Meta, GPU Arrays, Llama3
News Content: Meta recently disclosed details about its two newly-built large GPU clusters, which will be utilized for training its advanced Llama3 language model. In a blog post titled “Building Meta’s GenAI Infrastructure,” the company extensively outlined the hardware, network, storage, design, performance, and software configurations of these clusters, each composed of 24,576 Nvidia Tensor Core H100 GPUs. These new clusters represent an expansion and upgrade from the AI Research SuperCluster (RSC) launched in 2022.
The substantial increase in GPU count, from the previous 16,000 Nvidia A100 GPUs, provides Meta’s AI research and development with significantly greater computational power. The new GPU cluster design is geared towards supporting cutting-edge AI applications in natural language processing, speech recognition, and image generation. Meta claims that this scalability enables them to handle larger and more complex models, thus laying a solid foundation for advancements in generative AI technology.
The unveiling of these two state-of-the-art GPU clusters not only underscores Meta’s continued investment in AI infrastructure but also foreshadows substantial improvements in the potential for AI product development. With enhanced computing capabilities, Meta is poised to make further innovations in key areas such as language understanding and generation, as well as visual recognition, thereby pushing the boundaries of artificial intelligence technology.
【来源】https://www.datacenterdynamics.com/en/news/meta-reveals-details-of-two-new-24k-gpu-ai-clusters/
Views: 3