Summary:
- OCI Zettascale10 by Oracle is designed for large-scale generative AI use cases like training and running language models.
- Oracle introduced new capabilities in their OCI networking stack to help customers run workloads more efficiently and cost-effectively.
- The zettascale supercomputer by Oracle is built on Acceleron RoCE architecture and Nvidia AI infrastructure for breakthrough scale and improved performance.
Article:
Oracle Unveils OCI Zettascale10 for Large-Scale AI Workloads
Oracle has recently launched OCI Zettascale10, a cutting-edge solution designed to cater to large-scale generative AI use cases such as training and running language models. According to Info-Tech’s Palanichamy, the primary goal of Zettascale10 is to provide a seamless integration for these advanced AI applications.
New Capabilities in Oracle Acceleron Networking Stack
Additionally, Oracle has introduced new features in their OCI networking stack, including dedicated network fabrics, converged NICs, and host-level zero-trust packet routing. These enhancements aim to help customers run workloads more efficiently and cost-effectively, with the potential to double network and storage throughput while reducing latency and costs.
Breakthrough Performance with Zettascale Supercomputer
Oracle’s zettascale supercomputer leverages Acceleron RoCE architecture and Nvidia AI infrastructure to deliver breakthrough scale, minimal GPU-to-GPU latency, and improved price/performance ratios. The new architecture boasts a wide, shallow, and resilient fabric, taking advantage of modern GPU NICs to connect to multiple switches simultaneously on isolated network planes.
Moreover, the power-efficient optics and hyper-optimized density of the zettascale supercomputer allow for larger clusters to be deployed faster, with fewer stalls and checkpoint restarts. This optimized architecture ensures smoother traffic flow, even in the face of unstable or contested paths within the network.
In conclusion, Oracle’s OCI Zettascale10 and the new capabilities in the Acceleron networking stack showcase the company’s commitment to providing cutting-edge solutions for large-scale AI workloads, offering breakthrough performance and enhanced efficiency for customers in need of advanced computing solutions.
