NCA-AIIO Question 114
Select 3An organization is planning to virtualize its AI infrastructure using GPUs to support multiple workloads and users. Which key considerations should the organization focus on to ensure optimal performance and resource utilization?
- A
Ensure the hypervisor supports GPU virtualization technologies such as NVIDIA vGPU.
- B
Configure CPU overcommitment to maximize the number of virtual machines.
- C
Plan for GPU memory allocation per virtual machine to avoid resource contention.
- D
Use a GPU scheduler that prioritizes latency-sensitive workloads when required.
- E
Select physical GPUs with the highest possible clock speeds, regardless of workload requirements.
Show answer and explanation
Correct answers: A, C, D
Explanation
Virtualizing accelerated infrastructure requires careful planning to ensure GPUs are effectively shared and workloads run efficiently. Key considerations include ensuring the hypervisor supports GPU virtualization technologies, managing GPU memory allocation to prevent resource contention, and leveraging scheduling policies to prioritize critical workloads. Overcommitting CPU resources and focusing solely on GPU clock speeds are not optimal strategies for virtualized AI infrastructure.
- A. Correct.
Ensuring the hypervisor supports GPU virtualization technologies like NVIDIA vGPU is critical for enabling GPU sharing across multiple virtual machines. Without this support, GPUs cannot be virtualized effectively.
- B. Incorrect.
Configuring CPU overcommitment may lead to performance degradation and is not a key consideration specific to GPU virtualization. Proper CPU and GPU resource allocation should be balanced.
- C. Correct.
Planning GPU memory allocation per virtual machine is essential for avoiding resource contention and ensuring each virtual machine has enough GPU memory for its workload.
- D. Correct.
Using a GPU scheduler that prioritizes latency-sensitive workloads is important for optimizing performance, especially in environments with diverse workload requirements.
- E. Incorrect.
Selecting physical GPUs with the highest clock speeds is not always necessary. The choice of GPU should align with workload requirements, such as computational needs, memory capacity, and power efficiency.