Meta 近日宣布了其最新的 AI 基础设施建设，揭示了两个全新的 24k GPU 集群，这些集群将被用于训练其先进的 Llama3 大语言模型。这一消息在 Meta 的一篇名为“Building Meta’s GenAI Infrastructure”的博文中发布，详细阐述了这些集群的硬件配置、网络架构、存储解决方案、设计创新、性能指标以及配套软件。
这两个全新的数据中心规模的集群基于 Meta 的 AI Research SuperCluster (RSC)，后者于2022年推出，彰显了公司在人工智能领域的持续投入和技术创新。新集群各自包含24,576个Nvidia Tensor Core H100 GPU，相比于原先的16,000个Nvidia A100 GPU集群，规模显著扩大，性能得到大幅提升。
这一升级的GPU数量不仅意味着更强的计算能力，还标志着Meta在自然语言处理、语音识别和图像生成等AI研究和开发领域的能力得到了质的飞跃。Meta表示，这些增强的集群能够支持更大、更复杂的模型训练，为生成式AI技术的未来发展开辟了新的道路，预示着更智能、更高效的AI产品即将面世。
Meta的这一举措再次凸显了其在构建高效能AI基础设施上的领导地位，为全球AI研究和应用设立了新的标准。随着Llama3等模型在新集群上的训练，我们有望见证更多突破性的AI技术应用于日常生活中。
英语如下：
**News Title:** “Meta Constructs 24k GPU Superclusters to Accelerate Training of Llama3 Language Model”
**Keywords:** Meta, GPU Clusters, Llama3
**News Content:** Meta recently announced its latest AI infrastructure development, unveiling two new 24k GPU clusters that will be utilized for training its advanced Llama3 large language model. The news was disclosed in a Meta blog post titled “Building Meta’s GenAI Infrastructure,” which delves into the hardware specifications, network architecture, storage solutions, design innovations, performance metrics, and accompanying software of these clusters.
These two new data center-scale clusters are based on Meta’s AI Research SuperCluster (RSC), which was introduced in 2022, demonstrating the company’s ongoing investment and technological innovation in the AI domain. Each of these new clusters comprises 24,576 Nvidia Tensor Core H100 GPUs, a substantial expansion from the previous 16,000 Nvidia A100 GPU clusters, resulting in a significant boost in performance.
The升级的GPU count not only signifies increased computational power but also marks a qualitative leap in Meta’s capabilities in AI research and development, particularly in natural language processing, speech recognition, and image generation. Meta asserts that these enhanced clusters enable the training of larger and more complex models, paving the way for new frontiers in generative AI technology and预告着 the imminent arrival of smarter, more efficient AI products.
This move by Meta underscores its leadership in building high-performance AI infrastructure and sets new standards for global AI research and application. With the Llama3 model training on these new clusters, we can anticipate more groundbreaking AI technologies being applied to everyday life.
【来源】https://www.datacenterdynamics.com/en/news/meta-reveals-details-of-two-new-24k-gpu-ai-clusters/
Views: 1