Partner POV | Bringing Flexibility to AI Workloads with the New Cisco UCS C845A M8 Rack Server

This article was contributed. by our partner, Cisco, and written by Jeremy Foster, Senior Vice President & General Manager, Cisco Compute

We've engineered this groundbreaking addition to our AI server lineup to deliver enterprise AI workloads, providing innovative NVIDIA accelerated computing solutions that equip businesses to thrive in today's fast-paced digital environment. The UCS C845A M8 Rack Server is an ideal fit for enterprise data center environments because it's highly scalable, flexible, and customizable—delivering powerful AI capabilities to mainstream enterprise peripheral component interconnect express (PCIe) servers.

Imagine a healthcare analytics company starting its AI journey with the UCS C845A M8, initially using two GPUs to analyze patient data. As its models and data volume grow, the company scales up to eight GPUs, seamlessly enhancing capabilities to help improve patient outcomes without disrupting operations.

This can be done because the server is based on the NVIDIA MGX modular reference design, which supports two to eight NVIDIA PCIe GPUs (including NVIDIA H200 NVL, H100 NVL, or L40S GPUs), NVIDIA BlueField-3 SuperNICs and DPUs, and NVIDIA AI Enterprise software.

This platform is not just about raw power; it's about unlocking new possibilities and efficiencies that were previously unimaginable. It builds on the capabilities of our UCS C885A M8 Rack Server—an eight-way accelerated computing system based on the NVIDIA HGX platform—by delivering exceptional performance and flexibility across a wide range of enterprise AI workloads.

Built for AI performance, efficiency, and scale

As we continue our AI journey, this introduction marks a pivotal step. The UCS C845A M8 leverages the NVIDIA MGX modular reference design and is future-proofed to integrate next-gen PCIe GPUs without requiring a new platform. Cisco plans to offer configurations with additional GPUs as they become available. It also features a dense design, supporting up to eight GPUs in a compact 4RU chassis.

We've also improved the reference design while preserving the intent of the design. This includes enhanced power delivery, fewer printed circuit boards (PCBs), and better cable routing for optimal airflow and thermal management. Because of this, if a GPU were to fail, we make it faster and easier to replace it and then get back up and running. The system features E1.S solid state drives for local storage to increase storage density, improve thermal management, and provide high performance in a compact form factor.

From day 1, IT teams can manage the UCS C845A M8 through Cisco Intersight. This means you can have the same operational capabilities for traditional and AI workloads. At the same time, you also get the latest hardware compatibility recommendations, security alerts, and integrations with your existing tools, like ServiceNow.

In addition to delivering accelerated servers to address compute-intensive AI workloads, we are also offering AI PODs to help shorten the time required to achieve production-ready inferencing. Built on the foundation of Cisco Validated Designs (CVDs), AI PODs include NVIDIA AI Enterprise and provide customers with an established starting point, easily adaptable to meet their specific needs. These full-stack, pre-sized infrastructure bundles eliminate the guesswork from deploying AI inference solutions—from edge inferencing to large-scale clusters with NVIDIA accelerated computing. This means faster time to value, consistent performance, and reduced risk for AI projects.

An AI revolution for data scientists, AI engineers, and CTOs

The UCS C845A M8 Rack Server's advanced architecture is designed to enable AI innovation, offering the computational power needed for intensive tasks and the efficiency required for rapid deployment. By advancing beyond traditional capabilities, this new AI server enables faster deployment of AI applications, fostering innovation and helping position businesses as leaders in the AI domain.

The server's modular NVIDIA MGX design allows for flexible configurations and addresses a broad range of AI use cases.

Here are a few examples:

• Large enterprises with significant data processing needs, such as those in finance, healthcare, automotive, and manufacturing, can utilize this server for their extensive computational requirements.
• Research institutions and universities, particularly those focusing on AI and machine learning, can leverage this type of server to analyze large, complex datasets across various fields to better tune predictions, identify patterns, and gain valuable insights that might be near impossible with conventional methods.
• Cloud service providers can leverage AI for deep learning and inferencing to effectively deliver IT services and SaaS applications.
• Government agencies, particularly those involved in large-scale data analysis, may have needs for projects requiring advanced computational capabilities.

Organizations with significant data processing needs can further accelerate their inferencing and RAG/CAG workloads by leveraging the SuperNIC or NIC card options available, like the latest NVIDIA BlueField-3 SuperNIC or NVIDIA ConnectX-7 cards. This would also allow for more flexibility to right-size workloads while saving power and reducing cost.

In addition, depending on the model and number of GPUs configured, the UCS C845A M8 Rack Server is ideal for:

• GenAI training and fine-tuning
• High-performance computing (HPC)
• Data analytics and visualization
• Hyperscale cloud applications
• Design and simulation
• Language processing
• Conversational AI
• Graphics and rendering
• Virtual AI workstations

As we reflect on the energy and innovation showcased at Cisco Live Amsterdam, today's unveiling stands as a testament to our commitment to leading the charge in AI technology. The UCS C845A M8 Rack Server is designed to empower your business to navigate the complexities of a data-driven world with confidence and precision.

We'd like to join you on your journey as you drive your enterprise forward in this exciting new era of AI. The Cisco UCS C845A M8 ships in early Q3CY25, with orders opening in May 2025.