NCA-AIIO Question 116
Select 3An organization is planning to virtualize its GPU-accelerated infrastructure to support AI workloads. Which key considerations should they prioritize to ensure optimal performance and resource allocation?
- A
Ensure the hypervisor supports GPU virtualization technologies such as NVIDIA vGPU.
- B
Focus solely on scaling CPU resources, as GPU tasks are independent of CPU performance.
- C
Evaluate the workload requirements to determine the appropriate GPU profile allocation for each virtual machine.
- D
Disable GPU passthrough to avoid over-committing GPU resources across VMs.
- E
Implement high-speed networking and storage to prevent bottlenecks in data transfer for GPU workloads.
Show answer and explanation
Correct answers: A, C, E
Explanation
Virtualizing GPU-accelerated infrastructure requires careful planning to optimize resource utilization and performance. Key considerations include ensuring the hypervisor supports GPU virtualization technologies like NVIDIA vGPU, evaluating workload requirements for appropriate GPU profile allocation, and implementing high-speed networking and storage to avoid bottlenecks. Overlooking these factors can lead to degraded performance and inefficient resource usage.
- A. Correct.
Correct: The hypervisor must support NVIDIA vGPU or similar technologies to enable efficient virtualization of GPU resources.
- B. Incorrect.
Incorrect: While CPU resources are important, focusing solely on them without considering GPU and other resource dependencies will lead to suboptimal performance.
- C. Correct.
Correct: Understanding workload requirements and allocating appropriate GPU profiles ensures proper resource utilization and avoids over- or under-provisioning.
- D. Incorrect.
Incorrect: GPU passthrough can be useful for some scenarios, but disabling it entirely limits flexibility in resource allocation and could negatively impact certain workloads.
- E. Correct.
Correct: High-speed networking and storage are critical in avoiding bottlenecks, particularly for data-intensive AI workloads that rely on GPU acceleration.