NCA-AIIO Question 113
Select 3You are tasked with virtualizing an accelerated infrastructure that leverages NVIDIA GPUs for AI workloads. Which considerations are most critical to ensure optimal performance and resource utilization in this environment?
- A
Ensure the hypervisor supports GPU virtualization technologies such as NVIDIA vGPU.
- B
Allocate all GPU resources to a single virtual machine to avoid resource contention.
- C
Evaluate the compatibility of the GPU drivers with the virtualized environment and hypervisor.
- D
Configure proper isolation and resource quotas for shared GPU instances.
- E
Avoid using GPU passthrough to maintain flexibility across multiple workloads.
Show answer and explanation
Correct answers: A, C, D
Explanation
When virtualizing accelerated infrastructure with NVIDIA GPUs, the key considerations include ensuring that the hypervisor supports GPU virtualization technologies (e.g., NVIDIA vGPU), verifying driver compatibility, and implementing proper isolation and resource quotas for shared GPU instances. These measures help maintain performance and ensure optimal resource utilization. Allocating all GPU resources to a single VM or avoiding GPU passthrough entirely are not aligned with best practices for virtualized environments.
- A. Correct.
Ensuring the hypervisor supports GPU virtualization technologies like NVIDIA vGPU is essential for enabling efficient virtualization of GPU resources.
- B. Incorrect.
Allocating all GPU resources to a single virtual machine is generally not recommended as it defeats the purpose of virtualization, which is to share resources efficiently across multiple workloads.
- C. Correct.
Evaluating the compatibility of GPU drivers ensures that the hardware, virtualization stack, and software layers work seamlessly together without performance or stability issues.
- D. Correct.
Proper isolation and resource quotas for shared GPU instances are critical to prevent one workload from impacting the performance of others in a multi-tenant environment.
- E. Incorrect.
While GPU passthrough can be useful in some cases, avoiding it is not a critical consideration for virtualized environments. In fact, GPU passthrough can sometimes be beneficial for specific workloads requiring direct hardware access.