CLF-C02 Cloud Technology and Services Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  EC2 GPU instances (P and G family)

  Why it's wrong here

  EC2 GPU instances in the P and G families use powerful NVIDIA GPUs, which are excellent for parallel compute and can run many ML workloads effectively. However, they are general-purpose GPU hardware that consumes significant power and has a higher cost per inference or training step compared to purpose-built ML silicon. AWS's custom Inferentia (Inf) and Trainium (Trn) chips are architected specifically for ML operations, delivering better price/performance and lower latency for production ML workloads. Therefore, while GPUs are capable, they are not the best answer for cost-optimized, purpose-built ML acceleration.

• ✓

  EC2 Inf and Trn instances (AWS Inferentia and Trainium)

  Why this is correct

  EC2 Inf and Trn instances are built on AWS Inferentia and Trainium, which are custom-designed ASICs (application-specific integrated circuits) created by AWS exclusively for machine learning. Inferentia instances provide high-throughput, low-latency inference at a lower cost than GPUs, while Trainium instances are optimized for training large models with high efficiency. These chips integrate natively with major frameworks like TensorFlow and PyTorch, and they are the correct choice when the question emphasizes custom ML hardware, cost-effective scaling, and high performance for ML inference or training workloads.

• ✗

  EC2 Compute-optimized instances (C family)

  Why it's wrong here

  EC2 Compute-optimized instances in the C family are powered by high-frequency Intel or AMD CPUs and are designed for general compute-intensive applications such as batch processing, web servers, and scientific simulation. They lack any dedicated ML accelerators, so they can only run ML workloads using CPU instructions, which are far less efficient for the matrix multiplications and tensor operations that machine learning requires. While they can be used for lightweight or non-critical ML tasks, they are not intended for high-performance ML training or inference and would provide poor performance and higher total cost for this use case.

• ✗

  AWS Lambda with extended memory

  Why it's wrong here

  AWS Lambda with extended memory is a serverless function service that scales compute automatically, but the memory setting simply allocates more vCPUs from standard x86 or Arm processors. Lambda does not expose any GPU, Inferentia, or Trainium hardware, so even at the maximum memory limit, it remains CPU-only and cannot perform low-latency or high-throughput ML inference. It is designed for event-driven, short-running code, not for continuous or resource-heavy ML acceleration, making it unsuitable here.

Go deeper

Related to this question