Meta 近日公开了其最新的两个24k GPU集群的详细规格，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，该公司详述了这些大规模数据中心的硬件、网络、存储、设计、性能和软件配置。这两个新集群是基于2022年推出的AI Research SuperCluster (RSC)构建的，旨在推动人工智能在自然语言处理、语音识别和图像生成等领域的研究与开发。
每个新集群均配备了24,576个Nvidia Tensor Core H100 GPU，相比于先前拥有16,000个Nvidia A100 GPU的原始集群，GPU数量显著提升。这一升级意味着新集群能够处理更大、更复杂的模型，为生成式AI技术的创新奠定了坚实基础。Meta指出，这种扩展能力将加速AI产品的开发进程，特别是在语言理解和生成方面。
Meta的这一举措展示了其在人工智能基础设施领域的持续投入和领先技术实力。随着Llama3模型在更强大的硬件支持下进行训练，预计未来将带来更智能、更高效的AI解决方案，进一步推动全球AI研究的边界。
英语如下：
**News Title:** “Meta Builds Ultra-large GPU Cluster: 24k GPUs Power Llama3 Language Model Training”
**Keywords:** Meta, GPU Cluster, Llama3
**News Content:** Meta has recently disclosed the detailed specifications of its two new 24k GPU clusters, which are designed to train its advanced Llama3 large language model. In a blog post titled “Building Meta’s GenAI Infrastructure,” the company elaborated on the hardware, network, storage, design, performance, and software configurations of these massive data centers. These new clusters are based on the AI Research SuperCluster (RSC) launched in 2022, aiming to advance AI research and development in natural language processing, speech recognition, and image generation.
Each of the new clusters is equipped with 24,576 Nvidia Tensor Core H100 GPUs, a significant increase from the original cluster’s 16,000 Nvidia A100 GPUs. This upgrade enables the new clusters to handle larger and more complex models, thereby solidifying the foundation for innovations in generative AI technologies. Meta highlights that this scalability will accelerate the development process of AI products, particularly in language understanding and generation.
This move by Meta underscores its ongoing investment and leading technological prowess in AI infrastructure. With the Llama3 model training on this more powerful hardware, it is anticipated that it will yield smarter and more efficient AI solutions, pushing the boundaries of global AI research further.
【来源】https://www.datacenterdynamics.com/en/news/meta-reveals-details-of-two-new-24k-gpu-ai-clusters/
Views: 1