NCA-AIIO Question 118
Select 3You are tasked with virtualizing an AI workload on an accelerated infrastructure that includes NVIDIA GPUs. Which key considerations should you prioritize to ensure optimal performance and resource utilization?
- A
Ensure that the hypervisor supports NVIDIA vGPU technology.
- B
Allocate more GPU memory than the physical GPU can support to maximize workload capacity.
- C
Verify that the workloads are compatible with the selected NVIDIA GPU driver version.
- D
Plan for GPU resource oversubscription to maximize GPU utilization across multiple virtual machines.
- E
Configure proper NUMA alignment to minimize latency and maximize performance.
Show answer and explanation
Correct answers: A, C, E
Explanation
Virtualizing an accelerated infrastructure with NVIDIA GPUs requires careful consideration of hypervisor support, driver compatibility, and hardware configurations. Ensuring the hypervisor supports NVIDIA vGPU technology and configuring NUMA alignment are critical steps for optimizing resource allocation and minimizing latency. Verifying workloads are compatible with the selected NVIDIA GPU driver version ensures stable and optimal operation. Avoiding oversubscription of GPU resources and respecting the physical GPU memory limits are equally important to maintain performance for AI workloads.
- A. Correct.
Ensuring hypervisor support for NVIDIA vGPU technology is crucial because not all hypervisors support GPU virtualization, and this support is required to allocate GPU resources to virtual machines effectively.
- B. Incorrect.
Allocating more GPU memory than the physical GPU can support is not feasible and will lead to resource allocation failures or degraded performance, as physical GPU memory capacity is a hard limit.
- C. Correct.
Verifying workload compatibility with the selected NVIDIA GPU driver version ensures that the software and hardware work seamlessly, avoiding driver-related errors or performance issues.
- D. Incorrect.
GPU resource oversubscription can lead to performance degradation, especially for AI workloads that require consistent GPU performance. AI workloads benefit from dedicated GPU resources.
- E. Correct.
Configuring proper NUMA alignment is essential in virtualized environments to minimize latency between CPU, memory, and GPU, thus improving overall workload performance.