Question 970
A team has corrected this HPE compute solution issue: An HPE ProLiant AI node displays unexpectedly low inference speed even though the GPU utilization graph is almost empty. Which continuing practice would best prevent the same problem?
Explanation
Correct Answer
Profile the complete inference pipeline before buying more GPUs. This recurring assurance measure is tied to the diagnosed risk: The workload may be bottlenecked before the accelerator rather than constrained by GPU capacity. For the active incident, the corrective action remains: Measure CPU preprocessing, data access and GPU queueing to locate the limiting stage.
Incorrect Answers
Storage location and access tier have not been matched to service-level requirements. The option addresses that risk, not the one responsible for the present issue. The workload may be bottlenecked before the accelerator rather than constrained by GPU capacity. Profile the complete inference pipeline before buying more GPUs.
The prevention measure is relevant to this condition: Unreachable devices cannot reliably report or receive expected management baseline changes. Here the recurring weakness is: The workload may be bottlenecked before the accelerator rather than constrained by GPU capacity. Profile the complete inference pipeline before buying more GPUs.
The chosen placement offers no validated alternate capacity for service recovery. The selected safeguard helps with that exposure, but it leaves the recurrence risk here unaddressed. The workload may be bottlenecked before the accelerator rather than constrained by GPU capacity. Profile the complete inference pipeline before buying more GPUs.