AI Hardware Fundamentals: GPUs, CPUs, Memory, Networking and Storage
For IT staff, buyers, managers and newcomers who need to understand the physical layer of AI well enough to read a spec sheet, size a simple system and ask vendors the right questions.
Level
Basic
Study time
8 hours
Lessons
20 in 6 modules
Exam
20 questions
About this course
AI projects succeed or stall on hardware decisions that are often made by people who were never shown how the physical layer works. This course explains it from the ground up without assuming an engineering background. You will learn why modern AI is as much a memory and data-movement problem as a compute problem, how CPUs, GPUs and other accelerators differ, what is inside a GPU (cores, tensor units, high-bandwidth memory) and why precision formats such as BF16, FP8 and FP4 change what fits and how fast it runs. You will get an overview of the current accelerator lineup from NVIDIA, AMD, Intel and the cloud providers, reviewed as of October 2026.
The second half follows the system outward: PCIe cards versus SXM and OAM modules, 8-GPU servers and rack-scale systems, Ethernet versus InfiniBand, NVMe and parallel filesystems, and the power and liquid cooling that dense GPU racks demand. The course closes with practical skills: reading a spec sheet critically, sizing an inference server and a small training system with worked arithmetic, and structuring a cloud-versus-buy decision. It ends with an automated exam and a Wirewalk Institute certificate of completion and assessment.
What you will be able to do
Explain why AI workloads are limited by compute, memory bandwidth and data movement, and identify which limit dominates for training and for inference
Distinguish CPUs, GPUs and other AI accelerators, and describe the role of tensor units, HBM and precision formats in a GPU
Compare PCIe, SXM and OAM form factors, 8-GPU servers and rack-scale systems, and state their power and cooling implications
Describe how Ethernet, InfiniBand, NICs and switches connect GPUs, and why bandwidth and latency matter for multi-GPU work
Explain the roles of local NVMe, parallel filesystems and object storage in feeding GPUs and saving checkpoints
Read an accelerator or server spec sheet and spot the figures that are peak, sparse or otherwise not comparable
Size the GPU memory and approximate throughput for a simple inference server and a small training system using back-of-envelope arithmetic
Structure a cloud-versus-buy comparison using utilization, total cost and operational factors
Syllabus
6 modules · 20 lessons · about 8 hours
1
Module 1: Why AI Is a Hardware Problem
The three limits on AI performance and the main kinds of processors that address them.3 lessons
Compute, Memory Bandwidth and Data Movement: The Three Limits
Training Versus Inference: Different Jobs, Different Bottlenecks
CPUs, GPUs and Other Accelerators in Plain Terms
2
Module 2: Inside the GPU and Its Memory
What a GPU is made of, how numbers are stored, how memory is layered, and who makes accelerators today.4 lessons
GPU Anatomy: Cores, Tensor Units and HBM
Precision Formats: From FP32 to FP4
The Memory Hierarchy: Cache, DRAM, HBM and Capacity Versus Bandwidth
The Accelerator Lineup in 2026: NVIDIA, AMD, Intel and Cloud Chips
3
Module 3: Servers, Form Factors, Power and Cooling
How accelerators are packaged into cards, servers and racks, and what that means for the data center.3 lessons
PCIe Cards Versus SXM and OAM Modules
8-GPU Servers and Rack-Scale Systems
Power and Cooling: kW per Rack and Liquid Cooling
4
Module 4: Networking for Multi-GPU Work
How GPUs talk to each other inside a server and across a cluster.3 lessons
Scale-Up and Scale-Out: Why GPUs Need to Talk
Ethernet Versus InfiniBand: NICs, Switches and Fabrics
Bandwidth, Latency and Simple Fabric Design
5
Module 5: Storage for AI
The storage tiers that hold datasets, models and checkpoints, and how fast they need to be.3 lessons
Local NVMe SSDs and What They Are Good For
Parallel Filesystems and Object Storage
Feeding the GPUs: Data Loading and Checkpoint Throughput
6
Module 6: Specifying and Buying AI Hardware
Read spec sheets critically, size simple systems with arithmetic, and decide between cloud and owning.4 lessons
Reading an Accelerator and Server Spec Sheet
Sizing an Inference Server: A Worked Example
Sizing a Small Training System: A Worked Example
Cloud Versus Buying: Making the Call
Exam and certificate
The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 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, Intel, Google Cloud, AWS, Microsoft Azure, Meta, Dell, HPE, Supermicro, Lenovo, Arista, Cisco, Broadcom, IBM, DDN, WEKA, or VAST Data. Product and company names are trademarks of their respective owners.
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.
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.
How server CPUs are built and where they matter in AI systems: cores, chiplets, caches, memory channels and PCIe lanes; vector and matrix extensions from AVX-512 and AMX to SVE2 and SME; the 2026 Xeon 6, EPYC 9005 and 9006, Grace and Vera, and cloud Arm lineups; and how to pick and tune a host CPU so GPUs are never starved. Reviewed against vendor sources as of October 2026.