🗺️ Roadmap: The Physical Realm — Data Center Foundations & Hardware Architecture¶

Below is the syllabus timeline of lessons in this module. Track your study progress here.

📈 Progress¶

---

📁 1 What Is a Computer? Core Architecture for Absolute Beginners¶

• 1.1a The Fetch-Decode-Execute Cycle explained step by step
• 1.1b Memory bus, address bus, and data bus — the three communication highways
• 1.1c Why the Von Neumann bottleneck limits AI workloads (preview of GPU need)
• 1.2a Cores, threads, and clock speed: understanding GHz and parallelism
• 1.2b Cache hierarchy: L1, L2, L3 — why nearness to the core matters
• 1.2c CPU architectures for AI servers: Intel Xeon Scalable vs AMD EPYC
• 1.2d NUMA (Non-Uniform Memory Access): why socket locality matters for AI
• 1.2e SMT / Hyper-Threading: logical vs physical cores
• 1.3a DDR4 vs DDR5: generation differences, bandwidth, and latency
• 1.3b DIMM form factors: UDIMM, RDIMM, LRDIMM, and why AI servers need RDIMMs
• 1.3c ECC (Error-Correcting Code) Memory: why bit-flip errors can corrupt a 72-hour training run
• 1.3d Memory channels, ranks, and interleaving for maximum bandwidth
• 1.3e Persistent Memory (PMem / Optane): bridging storage and DRAM
• 1.4a HDD mechanics: platters, read/write heads, seek time, and rotational latency
• 1.4b SSD internals: NAND Flash cells (SLC, MLC, TLC, QLC) and wear leveling
• 1.4c NVMe over PCIe: why NVMe is 5–7x faster than SATA SSDs for AI data pipelines
• 1.4d U.2, U.3, M.2, and E1.S form factors — physical packaging differences
• 1.4e Endurance ratings (TBW) and enterprise vs consumer grade drives
• 1.5a PCIe generations: Gen 3 (8 GT/s), Gen 4 (16 GT/s), Gen 5 (32 GT/s), Gen 6 preview
• 1.5b Lane widths: x1, x4, x8, x16 and bandwidth calculations per generation
• 1.5c PCIe slots, risers, and switch chips: how multiple GPUs connect
• 1.5d PCIe bifurcation: splitting one x16 slot into multiple x8 or x4 links
• 1.5e The PCIe bottleneck: why GPU-to-GPU traffic through the CPU is catastrophically slow
• 1.6a Server form factors: Tower, 1U, 2U, 4U, and blade systems
• 1.6b Motherboard chipsets and platform controller hubs in enterprise systems
• 1.6c Redundant power supplies: N+1 and 2N configurations for zero downtime
• 1.6d Hot-swap components: drives, PSUs, and fans without powering down
• 1.6e BIOS vs UEFI: firmware differences and Secure Boot implications for Linux
• 1.7a Baseboard Management Controller (BMC): the always-on embedded controller
• 1.7b IPMI (Intelligent Platform Management Interface): protocol and ipmitool commands
• 1.7c Redfish API: the modern RESTful replacement for IPMI
• 1.7d Vendor implementations: iDRAC (Dell), iLO (HPE), XCC (Lenovo), NVIDIA BMC
• 1.7e Tasks via BMC: power cycling, boot order, sensor readings, KVM-over-IP

📁 2 Data Center Facility Operations¶

• 2.1a AC vs DC power: understanding voltage, amperage, and wattage (P=IV)
• 2.1b Single-phase vs three-phase power and why AI racks need three-phase
• 2.1c PDUs (Power Distribution Units): basic vs metered vs switched vs intelligent
• 2.1d UPS (Uninterruptible Power Supplies): online double-conversion vs line-interactive
• 2.1e Power Usage Effectiveness (PUE): the efficiency metric of a data center
• 2.1f Calculating TDP (Thermal Design Power) and rack wattage for GPU dense configurations
• 2.1g Generator backup systems and automatic transfer switches
• 2.2a Fundamentals of heat transfer: conduction, convection, and radiation in servers
• 2.2b Hot-aisle / cold-aisle containment: the standard data center airflow model
• 2.2c Computer Room Air Conditioners (CRACs) and Air Handlers (CRAHs)
• 2.2d In-row and overhead cooling systems for high-density deployments
• 2.2e Why air cooling hits a wall: TDP limits approaching 500W per GPU
• 2.2f Direct Liquid Cooling (DLC): cold plates, coolant distribution units (CDUs), and manifolds
• 2.2g Immersion cooling: single-phase vs two-phase dielectric fluid
• 2.2h Rear-door heat exchangers as a transitional solution
• 2.3a Rack Units (RU/U): 42U and 48U standard racks, rail systems, and cable management
• 2.3b Weight distribution: floor loading limits and seismic bracing for DGX systems
• 2.3c Structured cabling: Cat6A copper for 10GbE management, fiber for high-speed networks
• 2.3d Top-of-Rack (ToR) switches: how servers connect to the network fabric
• 2.3e Overhead cable trays, blanking panels, and airflow management best practices
• 2.3f Data center tiers: Tier I–IV (Uptime Institute) and availability percentages
• 2.4a Badge access systems, mantrap entries, and biometric authentication
• 2.4b CCTV, auditing, and chain-of-custody for hardware
• 2.4c Secure hardware decommissioning and drive destruction standards