Meta公司近日宣布，其人工智能研究超级集群（RSC）迎来重大升级，推出了两个新的24,576个GPU的计算集群，以支持其正在开发的先进语言模型Llama 3的训练。这一升级显著提升了RSC的计算能力，为Meta在人工智能领域的研究和发展提供了强大的支持。
Meta在其官方博客中详细介绍了这两个新的GPU集群的硬件配置、网络结构、存储系统、设计理念、性能指标以及相关的软件技术。这些集群基于Nvidia的Tensor Core H100 GPU构建，相较于之前的16,000个Nvidia A100 GPU的原始集群，计算能力的提升为开发更大、更复杂的AI模型提供了可能。
此次升级的GPU集群不仅能够处理自然语言处理任务，还能支持语音识别和图像生成等技术。Meta表示，这些集群的强大算力将推动生成式AI产品开发的进步，为未来的AI应用打开新的大门。
英文标题：Meta Unveils Details of Two New 24k GPU Clusters for Training Llama 3
英文关键词：Meta AI, GPU Clusters, Llama 3 Model
英文新闻内容：
Meta has recently announced the launch of two new GPU clusters with 24,576 units each, as part of its artificial intelligence research supercluster (RSC). These clusters are designed to support the training of Llama 3, an advanced language model being developed by the company.
In a detailed blog post, Meta has outlined the hardware specifications, network architecture, storage systems, design philosophy, performance metrics, and software technologies of the two new clusters. Built on Nvidia’s Tensor Core H100 GPUs, the upgrade significantly enhances the computing power of RSC, enabling the development of larger and more complex AI models.
The new GPU clusters are not only capable of handling natural language processing tasks but also support voice recognition and image generation technologies. Meta has stated that the increased computational power will facilitate advancements in generative AI product development, paving the way for future AI applications.
【来源】https://www.datacenterdynamics.com/en/news/meta-reveals-details-of-two-new-24k-gpu-ai-clusters/
Views: 0