Meta 近日公开了其两个全新的 24k GPU 集群的详细信息，这些集群将被用于训练其先进的 Llama3 大语言模型。在一篇名为“Building Meta’s GenAI Infrastructure”的博文中，该公司详述了这些集群的硬件、网络、存储、设计、性能和软件配置。这两个新的数据中心规模的集群基于 Meta 的 AI Research SuperCluster (RSC)，该集群于2022年推出，展示了Meta在人工智能基础设施建设上的持续创新。
新发布的GPU集群各拥有24,576个Nvidia Tensor Core H100 GPU，相比于RSC原始的16,000个Nvidia A100 GPU，这是一个显著的提升。这一扩展不仅增强了集群的计算能力，也使其能够处理更大、更复杂的AI模型。据Meta透露，这些增强的集群将支持自然语言处理、语音识别和图像生成等领域的前沿研究和开发工作。
这一升级的GPU集群为Meta的生成式AI产品的未来发展奠定了坚实的基础。随着硬件性能的提升，Meta有望在AI技术上取得更大的突破，进一步推动人工智能在实际应用中的边界。这一举措再次彰显了Meta在人工智能领域的雄心壮志，以及其致力于构建更高效、更强大的AI基础设施的决心。
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 has recently disclosed details about its two new 24k GPU clusters, which will be utilized 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 clusters. These new data center-scale clusters are based on Meta’s AI Research SuperCluster (RSC), which was introduced in 2022, demonstrating the company’s ongoing innovation in AI infrastructure.
Each of the newly released GPU clusters comprises 24,576 Nvidia Tensor Core H100 GPUs, a substantial increase from the RSC’s initial 16,000 Nvidia A100 GPUs. This expansion not only boosts the cluster’s computational power but also enables it to handle larger and more complex AI models. According to Meta, these enhanced clusters will support cutting-edge research and development in areas such as natural language processing, speech recognition, and image generation.
This upgraded GPU cluster solidifies the foundation for Meta’s future development of generative AI products. With the boost in hardware performance, Meta is poised to make significant advancements in AI technology, further pushing the boundaries of AI in practical applications. This move underscores Meta’s ambitious stance in the AI domain and its commitment to building more efficient and powerful AI infrastructure.
【来源】https://www.datacenterdynamics.com/en/news/meta-reveals-details-of-two-new-24k-gpu-ai-clusters/
Views: 1