Wirewalk Instituteof AI
hw-302 Expert Waitlist

Server and Rack Architecture for AI: HGX, MGX, Rack-Scale Systems and OCP

For senior infrastructure, hardware and data center engineers who design, procure and deploy AI servers and racks, from 8-GPU HGX nodes to liquid-cooled rack-scale systems and OCP open racks.

The curriculum is being prepared and reviewed. Enrollment is not open yet.

About this course

AI compute is now bought and deployed as racks rather than servers, and the hard problems have moved from the motherboard to the rack and the facility: 100 to 200 kW per rack and rising, liquid cooling loops, busbars and power shelves, cable cartridges and spine trays, floor loading and the commissioning sequence that brings it all up. This course gives senior engineers an end-to-end architectural model of AI servers and racks. It starts with the node: HGX baseboards, SXM and OAM modules, PCIe switch and NIC topologies, host CPU and memory, boot and management controllers, and the MGX modular reference architecture used by OEMs.

It then moves to rack-scale systems current in October 2026: NVIDIA GB200 and GB300 NVL72 and the Vera Rubin NVL72 that began shipping in 2026, the announced Kyber rack for Rubin Ultra, and AMD's Helios rack built on the OCP Open Rack Wide specification. It covers the OCP standards that shape the rack (Open Rack v3, Open Rack Wide, DC-MHS, OAM and the power and cooling workstreams), power delivery from AC and 48 V busbars to the 800 VDC architectures NVIDIA and partners are introducing, and direct liquid cooling with cold plates, manifolds, in-rack and in-row CDUs and facility water. Management with Redfish and OpenBMC, serviceability and spares, and a full planning and commissioning sequence close the course. Shipping status and specifications are stated as of October 2026, and announced or roadmap products are labeled. The course ends with an automated exam and a Wirewalk Institute certificate of completion and assessment; it is not a vendor certification.

What you will be able to do

  • Explain the architecture of an 8-GPU HGX node, including GPU baseboard, PCIe switch, NIC, CPU and management topology
  • Compare HGX, MGX, OEM and ODM server platforms, and OAM and SXM module ecosystems, for a stated deployment
  • Describe the architecture of rack-scale systems such as GB300 NVL72, Vera Rubin NVL72 and AMD Helios, including compute trays, switch trays, spine and power shelves
  • Apply OCP Open Rack v3, Open Rack Wide and related specifications to rack and facility design decisions
  • Design power delivery for an AI rack row, comparing AC, 48 V busbar and 800 VDC architectures on efficiency, current and safety
  • Specify a direct liquid cooling system including cold plates, manifolds, CDUs, facility water temperatures and leak detection
  • Plan procurement, site readiness, commissioning and serviceability for a multi-rack AI deployment

Syllabus

6 modules · 23 lessons · about 18 hours

Module 1: The AI Server Node

What is inside an 8-GPU server and the modular architectures used to build them.
4 lessons
  1. HGX Baseboards, SXM and OAM Modules
  2. Node Topology: PCIe Switches, NICs, Host CPUs and Storage
  3. MGX and Modular Server Architectures
  4. OEM Versus ODM Platforms: What You Actually Buy

Module 2: Rack-Scale Systems

NVL72-class racks, their compute and switch trays, and where rack-scale is heading.
4 lessons
  1. GB200 and GB300 NVL72: Compute Trays, NVLink Switch Trays and the Spine
  2. Vera Rubin NVL72 and the Kyber Roadmap
  3. AMD Helios and Open Rack Wide
  4. Rack-Scale Versus 8-GPU Nodes: Choosing the Unit of Deployment

Module 3: Open Compute Standards

The OCP specifications that define racks, servers and modules for AI.
3 lessons
  1. Open Rack v3 and Open Rack Wide: Mechanics, Busbars and Power Shelves
  2. DC-MHS, OAM and Modular Hardware Specifications
  3. Following OCP Work in Progress: Power, Cooling and Resilience Workstreams

Module 4: Power Delivery

From utility to GPU: rack power architectures and their trade-offs.
4 lessons
  1. Rack Power Budgets: From 40 kW to Several Hundred kW
  2. 48 V Busbars, Power Shelves and Battery Backup Units
  3. 800 VDC and High-Voltage DC Architectures
  4. Power Capping, Transients and Grid Interaction

Module 5: Liquid Cooling

Direct liquid cooling design from cold plate to facility water.
4 lessons
  1. Cold Plates, Manifolds and Quick Disconnects
  2. CDUs: In-Rack, In-Row and Facility Loops
  3. Facility Water Temperatures, Heat Reuse and Hybrid Air Cooling
  4. Leak Detection, Water Quality and Liquid Cooling Operations

Module 6: Management, Deployment and Commissioning

Out-of-band management, site readiness, commissioning and serviceability at rack scale.
4 lessons
  1. BMCs, Redfish and OpenBMC for AI Racks
  2. Site Readiness: Floor Loading, Clearances, Power and Water
  3. Commissioning Sequence and Acceptance Criteria
  4. Serviceability, Spares and Field Replacement at Rack Scale

Exam and certificate

The exam unlocks once you have completed every lesson. It draws 25 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 55 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.

Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.

The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by NVIDIA, AMD, Open Compute Project, Dell Technologies, HPE, Supermicro, Lenovo, Vertiv, Schneider Electric, or DMTF. Product and company names are trademarks of their respective owners.

hw-001 Free intro Available

Inside an AI Data Center: GPUs, Networks and Storage in One Hour

A plain-English, one-hour walk through an AI data center: what a model asks of hardware, what is inside a GPU server, how GPUs talk over NVLink, InfiniBand and Ethernet, where the data lives, and why AI racks need so much power and cooling.

1 h · 6 lessons Free intro
hw-101 Basic Available

AI Hardware Fundamentals: GPUs, CPUs, Memory, Networking and Storage

A plain-language tour of the hardware that runs AI: why AI is limited by compute, memory bandwidth and data movement; what GPUs, HBM and precision formats are; how servers, networks, storage, power and cooling fit together; and how to size a simple inference or training system and weigh cloud against buying.

8 h · 20 lessons See price at checkout
hw-201 Intermediate Available

GPU Hardware Deep Dive: Architecture, Memory and Interconnects

An engineer's tour of the modern data center GPU: how SMs and CUs, warps and wavefronts, tensor and matrix cores and the on-chip memory hierarchy work; what FP64 through FP4 and sparsity really buy you; how HBM, PCIe, NVLink, Infinity Fabric, UALink and CXL move data; and how power, cooling and reliability features shape real deployments. Reviewed against vendor sources as of October 2026.

14 h · 21 lessons See price at checkout