Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore Cisco's comprehensive approach to AI infrastructure through this 31-minute conference presentation that addresses the critical challenges of GPU cluster design, deployment automation, and end-to-end visibility. Learn how Cisco simplifies AI infrastructure to combat lengthy design cycles and troubleshooting issues that can underutilize expensive GPU resources. Discover Cisco's three-pillar AI networking strategy, starting with custom Silicon One platforms featuring programmable pipelines that adapt to evolving AI demands, NVIDIA partnerships for full-stack reference architectures, and mature NX-OS software optimized for AI workloads. Examine the operating model encompassing Nexus Dashboard for on-premises management, Nexus Hyperfabric for cloud-managed solutions, and API-first approaches for seamless automation integration. Understand extensive AI reference architectures that serve as validated blueprints for deployments ranging from enterprise-scale (under 1024 GPUs) to hyperscale environments (1K-16K+ GPUs), with detailed component specifications ensuring consistent networking experiences across NVIDIA, AMD, and storage vendors. Master the design principles for efficient AI backend networks using non-blocking architectures that maintain 1:1 subscription ratios, keeping every GPU within one hop for optimal communication. Analyze Cisco's "scalable unit" concept for incremental expansion and policy-based load balancing innovations that prioritize critical GPU-to-GPU training traffic over storage and front-end communications. See practical implementation through a 32-GPU university cluster example demonstrating network convergence strategies and upcoming Nexus Dashboard platform updates for streamlined design, automation, and monitoring at scale.
