Oracle Cloud Infrastructure has unveiled its Zettascale10 supercomputer, integrating hundreds of thousands of Nvidia GPUs to deliver unprecedented processing power. It boasts capabilities reaching up to 16 zettaFLOPS, illustrating the architecture’s potential for handling complex AI and machine learning tasks in real-time. Analyst Yaz Palanichamy emphasized that this platform is designed for accelerated performance and operational efficiency tailored to large-scale AI applications, especially for training and executing large language models (LLMs).
Further, enhancements to Oracle’s Acceleron networking stack have introduced features for faster workload execution at reduced costs, doubling network throughput while minimizing latency. Oracle’s architecture features an innovative structure designed for resilience and efficiency, enabling rapid deployment of larger clusters with minimal downtime. However, concerns about the resource requirements and supply chain stability for such extensive AI operations remain a critical focus, as highlighted by analysts discussing the reliance on strategic partnerships for GPU demand stabilization. Palanichamy viewed Oracle’s collaboration with AMD positively, indicating significant progress towards energy efficiency in large-scale AI processing.
👉 Pročitaj original: CIO Magazine
