BlogCiscoIntelNVIDIA
Blog • • 8 minute read

Powering AI at Scale: Cisco UCS C880A M8

Cisco and World Wide Technology unveil the UCS C880A M8, an air-cooled, 8-way GPU server designed for large-scale AI workloads. This innovation enhances Cisco's AI infrastructure, offering exceptional GPU density, high-speed interconnects, and architectural flexibility. It represents a strategic evolution in secure, scalable, and automated AI operations, bridging current needs with future hybrid cooling solutions.


Powering AI at Scale 

Cisco and World Wide Technology continue their partnership in AI infrastructure innovation with the introduction of the Cisco UCS C880A M8, a next-generation air-cooled 8-way GPU server engineered for large-scale AI training, fine-tuning, and high-performance computing workloads. The system expands Cisco's Unified Computing System portfolio with exceptional GPU density, high-speed interconnects, and architectural flexibility for organizations advancing their AI strategy. More than a technical milestone, the C880A M8 represents a strategic evolution in enterprise AI infrastructure as a core building block of the Cisco Secure AI Factory with NVIDIA, a full-stack framework that unites Cisco's Enterprise Reference Architecture (ERA) and Cloud Reference Architecture (CRA) to deliver consistent, secure, and automated AI operations from core to cloud.

At the foundation of this design is Cisco's SP4 Switching platform and SPX Licensing, which enable high-bandwidth, multi-plane GPU networking, real-time telemetry, and policy-driven fabric orchestration across thousands of GPUs. Through SPX, customers can activate advanced capabilities such as secure multi-tenant partitioning, workload-aware energy optimization, and integrated observability through Cisco Intersight. When paired with Run:AI, the C880A M8 extends these capabilities into the software layer, providing intelligent GPU scheduling, dynamic resource pooling, and workload prioritization that help maximize utilization and accelerate time to insight.

By combining Cisco's full-stack architecture from SP4 switching and SPX fabric control to unified compute under ERA and CRA blueprints, the C880A M8 delivers supercomputing-class performance with enterprise-grade governance, sustainability, and operational simplicity. It reflects Cisco and WWT's shared vision for the next generation of AI infrastructure that is secure, scalable, and intelligently automated from silicon to service.

Built for AI at scale

At its core, the UCS C880A M8 leverages the NVIDIA HGX™ B300 platform featuring eight NVIDIA Blackwell Ultra GPUs interconnected through NVIDIA NVLink and NVLink Switch technology to deliver up to 1.8 TB/s of GPU-to-GPU bandwidth within a single node. Each GPU is paired with a NVIDIA  ConnectX-8 NIC providing up to 800 Gb/s of east-west bandwidth for multi-node AI clusters. This architecture enables highly efficient data parallel training and low-latency communication essential for modern AI workloads.

 

The system is powered by dual 6th Gen Intel Xeon Scalable Processors and high-speed DDR5 memory, providing balanced CPU and GPU performance for complex data pipelines. With PCIe Gen 5 connectivity, E1.S NVMe storage options, and a modular air-cooled chassis design, the C880A M8 delivers scalability and serviceability ideal for organizations expanding AI infrastructure without requiring liquid cooling.
 


The CPU tray design shown above highlights the C880A M8's modular architecture and balanced compute-to-I/O ratio.

  • PCIe Gen5 x16 slots accommodate high-bandwidth devices for data movement or specialized accelerators.
  • E1.S NVMe SSDs deliver high-speed local data access optimized for AI training pipelines.
  • Dual M.2 RAID boot drives provide reliability for OS and hypervisor environments.
  • X710 OCP NIC supports out-of-band management and telemetry through Cisco Intersight.
  • DC-SCM integration enhances system security with hardware root-of-trust and isolated firmware control.

This tray-level modularity allows easy serviceability while maintaining peak data throughput and I/O consistency, forming the backbone of the C880A M8's compute subsystem.
 

The table above illustrates available UCS C880A M8 configurations, highlighting component and GPU alignment across NVIDIA HGX™ B300-series builds. Each configuration is optimized for a specific performance or workload profile, offering flexibility to balance GPU density, memory capacity, and network bandwidth based on deployment needs.

Network Fabric Innovation

The UCS C880A M8 introduces dual-plane and quad-plane interconnect options to support multi-fabric GPU topologies with active-active redundancy and load-balanced throughput. Each plane functions as an independent network fabric, ensuring non-blocking, rail-optimized connectivity across large GPU clusters.

Dual-plane and four-plane design overview:

Dual-plane design: 2x 400GE per GPU for inter-GPU backend connectivity across eight NVIDIA Blackwell Ultra GPUs with two NVLink Switches, delivering balanced performance and redundancy for mid-scale distributed AI clusters.

Four-plane design: 4x 200GE per GPU, doubling link density and providing enhanced scalability for large-scale AI training or multi-tenant environments that require higher aggregate bandwidth and resiliency.
 

Typical deployments include 256-GPU (32 nodes) and 512-GPU (64 nodes) configurations using Cisco Nexus 9364E-SG2 400GE switches in a two-stage spine-leaf topology. This design supports both backend GPU-to-GPU training networks and frontend fabrics for inference, data ingestion, and management traffic. The result is simplified scaling, reduced latency, and balanced throughput without the need for additional super-spine layers.

Inter4-GPU Backend Network

Example applications:

  • Large-scale AI model training: Provides ultra-low latency and massive bandwidth to synchronize thousands of GPUs efficiently, reducing training time for LLMs and multimodal systems.
  • AI inference and production: Maintains predictable throughput and high availability, even in the event of a plane or link failure.
  • High-performance computing (HPC): Powers parallel simulations and data-intensive workloads such as molecular dynamics, climate modeling, and astrophysics research.
  • Data analytics and visualization: Accelerates distributed analytics pipelines with frameworks like RAPIDS and Apache Spark.
  • Multi-tenant environments: Enables workload segmentation by assigning network planes to different traffic types for isolation and scalability.
  • Research and validation: Supports benchmarking, model optimization, and topology testing within WWT's AI Proving Ground before production deployment.

Together, these capabilities make the UCS C880A M8 network fabric the foundation of a scalable and resilient AI infrastructure, providing the flexibility, bandwidth, and reliability needed to power the most demanding enterprise and research workloads.

Power, Cooling, and Security

The C880A M8 is engineered for efficient, air-cooled operation that maintains high performance under sustained GPU workloads while simplifying deployment and reducing cost. Air cooling remains one of the most reliable and energy-efficient methods for modern data centers, eliminating the need for liquid loops or facility retrofits while minimizing maintenance complexity. Advanced airflow design and thermal zoning ensure consistent cooling and stable GPU performance within standard rack environments.

Key advantages of air-cooled design:

  • Lower complexity and cost: Uses existing HVAC and electrical systems without specialized plumbing.
  • Operational flexibility: Supports hybrid environments that combine traditional compute and dense AI workloads.
  • Scalable path forward: Can integrate containment systems or rear-door heat exchangers as workloads grow.

As AI density increases, the future of data center cooling will evolve toward hybrid models that blend air, liquid, and immersion technologies based on workload intensity. Cisco's roadmap already anticipates this evolution, with systems like the UCS C885A M8 incorporating liquid-assisted cooling for higher thermal design power GPUs. The C880A M8 bridges this transition, offering efficient air-cooled operation today while preparing organizations for future hybrid cooling adoption.

Security and manageability:

  • Modular DC-SCM architecture integrates root of trust, BIOS, and BMC functions for secure boot and firmware integrity.
  • Dedicated management networks isolate administrative access and enhance telemetry visibility.
  • FIPS-compliant storage options ensure readiness for government, defense, and regulated industries.

Together, these features deliver a secure, efficient, and future-ready platform that aligns with Cisco's sustainability and security standards.

Workload Versatility and Use Cases

The UCS C880A M8 merges technical excellence with enterprise practicality, empowering organizations to deploy AI at scale without increasing operational risk. It supports diverse workloads—from model training to data analytics—making it a strategic investment for digital transformation.

Example use cases include:

  • Financial services: Fraud detection, risk analysis, and algorithmic trading.
  • Healthcare and life sciences: Diagnostic imaging, genomics analysis, and AI-assisted drug discovery.
  • Manufacturing and logistics: Predictive maintenance, automated inspection, and supply chain optimization.
  • Retail and consumer industries: Recommendation engines, demand forecasting, and customer behavior analysis.
  • Scientific computing: Climate modeling, molecular simulation, and astrophysical research.
  • Digital twin systems: Modeling smart factories, autonomous vehicles, and energy grids for real-time optimization.

With built-in security, flexible scaling, and compatibility with leading AI and analytics frameworks, the C880A M8 turns infrastructure into a catalyst for innovation.

Integration and Manageability

As part of Cisco's Unified Computing System, the C880A M8 integrates seamlessly with Cisco Intersight for centralized lifecycle management, policy-based automation, and real-time telemetry. This provides a unified operational model across distributed environments, simplifying management while maintaining full-stack visibility.

The C880A M8 also supports integrations with Splunk Observability Cloud, ThousandEyes, and Intersight Workload Optimizer, enabling real-time performance analytics and predictive capacity planning. Compatibility with NVIDIA AI Enterprise software and Cisco's HyperFabric AI ecosystem ensures interoperability with NVIDIA NIM™ microservices, data pipelines, and enterprise MLOps frameworks.

Validated Through WWT's ATC and AI Proving Ground

Customers can evaluate the C880A M8 through WWT's Advanced Technology Center (ATC) and AI Proving Ground (AIPG), which offer production-ready environments for benchmarking and validation. These facilities allow organizations to simulate real-world workloads, test scaling strategies, and refine configurations before deployment.

By working with WWT's engineers and architects, customers gain access to proven design guidance, hands-on integration experience, and visibility into how the C880A M8 performs under realistic AI, HPC, and data-intensive conditions.

 

The Next Evolution in AI Infrastructure

The Cisco UCS C880A M8 represents the next step in enterprise AI computing, delivering high performance, efficiency, and scalability in an air-cooled, data center-ready form factor. It enables organizations to expand incrementally, aligning infrastructure investments with business growth while maintaining operational simplicity.

Validated within WWT's ATC and AIPG environments, the C880A M8 allows enterprises to benchmark and refine their AI strategies with confidence. For customers prioritizing efficiency, modular growth, and long-term scalability, the C880A M8 is the optimal choice. For production-scale environments that demand maximum GPU density, throughput, and liquid-assisted cooling, the C885A M8 offers a powerful complement.

Together, these two systems form the foundation of Cisco's AI compute portfolio for the Cisco Secure AI Factory with NVIDIA, providing organizations with a flexible, future-ready path to scale innovation securely and sustainably and from early experimentation to full production.

Technologies