#None #Cisco #Internet2 #AI Pods

Enhancing Research Capacity Through AI

In a significant development for the research community, Internet2 has announced its latest initiative to bolster the National Research Platform (NRP) by teaming up with Cisco. This collaboration is marked by the introduction of new AI PODs at the University of Nebraska–Lincoln and Sacramento State, designed to raise the available computational capacity for researchers.

As the research landscape evolves, the reliance on artificial intelligence becomes increasingly apparent. The capabilities of AI in transforming research methods are unparalleled, and the NRP is steadfast in its commitment to expanding shared computing facilities that support both researchers and students nationwide. The newly implemented AI PODs serve as an essential addition to the existing resources, enhancing access to GPU capacity when local systems reach their limits.

What Are AI PODs?

The Cisco AI PODs represent a systematic approach to augmenting the NRP’s infrastructure. By integrating pre-validated computing, networking, and management components, these PODs streamline the process of incorporating additional resources. This not only eases the engineering burden but also facilitates a consistent model for contributors to the NRP, allowing for scalability as the demand for AI computing rises.

In an age where research computing is becoming increasingly intricate, the advent of AI represents both challenges and opportunities. Institutions often face budget constraints while trying to meet growing demands, creating a scenario where a shared model can significantly alleviate access barriers. Supported by the National Science Foundation and over seventy organizations, the NRP operates as a community-driven platform, enabling institutions to pool resources and provide researchers access to advanced computational tools beyond their immediate capabilities.

Innovative Structure of the NRP

Rather than relying on a single, centralized supercomputer, the NRP efficiently accumulates GPU and CPU resources provided by various institutions, forming a cohesive Kubernetes-based fabric known as Nautilus. However, integrating compatible GPU nodes into this shared, multi-regional cluster can often pose challenges due to differing compatibility in compute, networking, and management standards. This is where Cisco AI PODs come into play: they address these integration challenges effectively.

Each AI POD comprises a modular design, featuring Cisco UCS servers and GPU nodes that are interconnected via high-bandwidth, low-latency Nexus networking. Cisco Intersight is employed for centralized management, simplifying the deployment of additional nodes into the NRP. The result is minimized risk and reduced setup time for institutions looking to contribute to the platform.

James Deaton, vice president of Network Services and Research Engagement at Internet2, emphasized the value these AI PODs provide: "Today, these two new AI PODs grant researchers additional GPU capacity when local resources fall short. Internet2’s role is pivotal in connecting the essential components needed for the NRP: a Cisco solution and the campuses prepared to host it. Strategic partnerships with industry players amplify our investment in the research and education communities."

Details of the New AI PODs

Together, the AI PODs at these two campuses deliver an impressive array of resources: 8 NVIDIA H200 GPUs boasting a combined 1.1 TB of memory, 128 CPU cores, 2 TB of system memory, and 90 TB of NVMe flash storage for staging data. Each site hosts a Cisco UCS C845A M8 server chassis, with its GPUs interconnected over NVLink to operate as a single pool of shared GPU memory.

Looking ahead, Internet2 remains dedicated to enhancing shared research infrastructure to ensure that researchers across the U.S. can access necessary computational resources. The complexities of data management and accessibility must also be addressed, underscoring the importance of reliable pathways between storage and compute.

Internet2’s existing national network supports dedicated Open Science Data Federation nodes in several major cities, facilitating greater proximity between datasets and computing resources. Moreover, 41 performance assurance service nodes have been deployed to provide valuable insights into the performance metrics across the NRP, effectively monitoring and troubleshooting issues as they arise.

The Future of Research with AI

The deployment of AI PODs is a testament to how strategic collaborations can foster considerable advancements in the research sector. Internet2's partnership with Cisco exemplifies how shared resources can drive collective progress, ultimately benefiting a wider educational and research community. As the infrastructure for research continues to adopt more AI capabilities, such partnerships and innovations will shape the landscape of research and education for years to come.

For more information about Internet2 and its initiatives, visit Internet2's official website.