Describe Solutions to Maximize Performance in AI Workloads CERTIFIED-AI-EXPERT Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A startup runs a batch inference workload that processes 10,000 images per hour. The workload is CPU-bound and currently runs on a single VM with 8 vCPUs. Demand spikes to 40,000 images per hour for a few hours each day. The team wants to handle the spike without over-provisioning the rest of the day. Which approach best meets the requirement?

   • AScale horizontally by adding more VMs during the spike and removing them afterward.
   • BMove the workload to a single GPU VM to accelerate processing.
   • CKeep the 8-vCPU VM and queue the excess images for overnight processing.
   • DVertically scale the single VM to 32 vCPUs permanently.
   Reveal Solution Discussion Select an answer first
2. 2foundation · medium

   When optimizing an AI workload, what is a typical trade-off between using a larger batch size and model accuracy?

   • ALarger batch sizes can speed up training but may reduce model accuracy.
   • BLarger batch sizes always improve accuracy.
   • CLarger batch sizes have no effect on accuracy.
   • DLarger batch sizes reduce training speed.
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   A team is training a model on 32 GPUs. GPU utilization is high (90%), but the total training time is longer than expected. Profiling shows that the GPUs are busy, but the loss is not decreasing as fast as expected. The team suspects the effective batch size is too large, causing convergence issues. Which change would most likely improve convergence without sacrificing GPU utilization?

   • ASwitch to a smaller model to speed up each iteration.
   • BReduce the learning rate proportionally to the increase in batch size.
   • CIncrease the batch size further to improve GPU utilization.
   • DReduce the number of GPUs to lower the effective batch size.
   Reveal Solution Discussion Select an answer first
5. 4foundation · medium

   Which interconnect is specifically designed for high-speed, low-latency communication between GPUs in a cluster?

   • AInfiniBand
   • BUSB
   • CWi-Fi
   • DEthernet (1 Gbps)
   Reveal Solution Discussion Select an answer first
6. 5foundation · medium

   What is the primary benefit of mixed precision training for GPU utilization?

   • AIt guarantees higher model accuracy.
   • BIt eliminates the need for gradient accumulation.
   • CIt reduces the number of training epochs needed.
   • DIt reduces the memory footprint and can increase compute throughput.
   Reveal Solution Discussion Select an answer first