**Meta发布全新24k GPU集群，加速Llama3大语言模型训练**
近日，全球科技巨头Meta在一篇官方博文中宣布，已建成两个全新的24k GPU集群，这些集群将被用于训练其先进的Llama3大语言模型。这一举措进一步展现了Meta在人工智能基础设施建设上的决心和实力。
这两个集群基于Meta的AI Research SuperCluster（RSC），该RSC于2022年推出，是全球领先的人工智能研究平台。新的训练集群在规模上显著提升，每个集群都配备了24,576个Nvidia Tensor Core H100 GPU，相比RSC原有的16,000个Nvidia A100 GPU，GPU数量增长了53%。
这一硬件升级的目的是支持更大型、更复杂的AI模型训练，特别是在自然语言处理、语音识别和图像生成等领域。Meta表示，这些强大的GPU集群将极大地推动生成式AI产品的研发，为未来AI技术的进步打开新的可能。
Meta的这一创新举措不仅强化了其在人工智能领域的领先地位，也为全球AI研究和开发树立了新的标杆。随着硬件性能的提升，可以预见，Llama3模型将能够处理更大数据集，生成更精准、更复杂的语言输出，从而在人工智能应用中带来更优质的服务和体验。
英语如下：
**News Title:** “Meta Constructs 24k GPU Superclusters to Accelerate Llama3 Language Model Training”
**Keywords:** Meta, GPU clusters, Llama3
**News Content:**
Meta, the global tech giant, recently announced in an official blog post the establishment of two cutting-edge 24k GPU clusters, which will be dedicated to accelerating the training of its advanced Llama3 large language model. This move underscores Meta’s commitment and prowess in AI infrastructure development.
Built upon Meta’s AI Research SuperCluster (RSC), introduced in 2022 as a leading global platform for AI research, these new training clusters represent a significant expansion. Each cluster is equipped with 24,576 Nvidia Tensor Core H100 GPUs, a 53% increase from the original RSC’s 16,000 Nvidia A100 GPUs.
The hardware upgrade is designed to accommodate the training of larger and more complex AI models, particularly in natural language processing, speech recognition, and image generation. Meta claims that these powerful GPU clusters will greatly advance the development of generative AI products, opening up new possibilities for the progression of AI technology.
This innovative step not only reinforces Meta’s leadership in the AI domain but also sets a new benchmark for global AI research and development. With the boost in hardware capabilities, it is anticipated that the Llama3 model will be able to handle larger datasets, generating more precise and complex language outputs, thereby enhancing the quality and experience of AI-driven services and applications.
【来源】https://www.datacenterdynamics.com/en/news/meta-reveals-details-of-two-new-24k-gpu-ai-clusters/
Views: 3