37.2a Mock Exam 1: Foundational focus — infrastructure, hardware, and networking¶

🧭 Context Introduction¶

Welcome to your first mock exam in the NVIDIA-Certified Associate: AI Infrastructure and Operations track. This assessment is designed for new engineers who are building foundational knowledge in AI infrastructure. The focus here is on hardware components, networking fundamentals, and infrastructure concepts that underpin modern AI workloads. Think of this as a warm-up — it tests your understanding of the physical and logical building blocks before diving into more complex operational scenarios.

⚙️ Exam Overview¶

• Format: 20 multiple-choice questions
• Time Limit: 30 minutes
• Focus Areas: GPU hardware, server architectures, network topologies, and storage basics
• Target Audience: Engineers new to AI infrastructure and operations
• Passing Score: 70% (14 out of 20 correct)

🛠️ Key Topics Covered¶

🔲 GPU Hardware Fundamentals¶

• Understanding NVIDIA GPU models (A100, H100, V100)
• GPU memory types (HBM2e, HBM3) and capacity
• NVLink and NVSwitch interconnects
• Thermal design power (TDP) and cooling requirements

🖥️ Server and Rack Architecture¶

• DGX systems (DGX A100, DGX H100)
• Standard rack configurations (42U, 48U)
• Power distribution units (PDUs) and redundant power supplies
• Cable management and airflow considerations

🌐 Networking Essentials¶

• InfiniBand vs. Ethernet for AI workloads
• Network interface cards (NICs) and smartNICs
• Topologies: Fat-tree, Dragonfly, and Torus
• Latency and bandwidth requirements for distributed training

💾 Storage Concepts¶

• NVMe vs. SATA SSDs
• Parallel file systems (Lustre, GPUDirect Storage)
• Storage tiers: hot, warm, cold
• Data ingestion pipelines

📊 Comparison Table: GPU Models for AI¶

📊 Visual Representation: Mock Exam Foundations validation¶

This diagram maps mock exam structures, auditing understanding of hardware specs and basic server configurations.

🕵️ Sample Exam Questions (With Explanations)¶

Question 1¶

Which GPU interconnect technology allows direct GPU-to-GPU communication without going through the CPU?

• A) PCIe Gen 4
• B) NVLink
• C) Ethernet
• D) SATA

✅ Correct Answer: B) NVLink

Explanation: NVLink is NVIDIA's high-speed direct GPU-to-GPU interconnect. It bypasses the CPU and PCIe bus, enabling faster data transfer between GPUs — critical for multi-GPU training.

Question 2¶

In a standard 42U rack, how many DGX H100 systems can typically be installed when accounting for power and cooling constraints?

• A) 8
• B) 4
• C) 16
• D) 2

✅ Correct Answer: B) 4

Explanation: Each DGX H100 system occupies approximately 6U of rack space and requires significant power (up to 10.2 kW). With power and cooling limits, most data centers can safely install 4 DGX H100 systems per 42U rack.

Question 3¶

Which network topology is most commonly used in large-scale AI clusters for its fault tolerance and scalability?

• A) Star topology
• B) Bus topology
• C) Fat-tree topology
• D) Ring topology

✅ Correct Answer: C) Fat-tree topology

Explanation: Fat-tree topology provides multiple redundant paths between nodes, high bisection bandwidth, and scales well for distributed AI training. It's the preferred choice for clusters using InfiniBand or high-speed Ethernet.

📋 Study Tips for New Engineers¶

• Focus on hardware specifications: Memorize key specs for A100, H100, and V100 GPUs — memory size, bandwidth, and interconnect type.
• Understand networking basics: Know the difference between InfiniBand (low latency, RDMA) and Ethernet (general purpose, higher latency).
• Practice with rack layouts: Visualize how GPUs, CPUs, NICs, and storage fit into a standard rack.
• Review power and cooling: AI infrastructure generates significant heat — understand TDP and airflow requirements.
• Use NVIDIA documentation: The official NVIDIA DGX and GPU datasheets are your best friends.

✅ Final Checklist Before the Exam¶

• [ ] I can name three NVIDIA GPU models and their memory capacities.
• [ ] I understand the role of NVLink in multi-GPU communication.
• [ ] I can explain the difference between InfiniBand and Ethernet.
• [ ] I know what a fat-tree topology looks like.
• [ ] I can calculate basic power requirements for a DGX system.
• [ ] I am familiar with NVMe and GPUDirect Storage concepts.

🚀 Next Steps¶

After completing this mock exam, review your incorrect answers and revisit the corresponding topics. Then move on to Mock Exam 2 (Intermediate focus — software, orchestration, and monitoring) to build on this foundation.

Good luck, and remember — AI infrastructure is about connecting the right hardware with the right network to make GPUs work efficiently together. You've got this!