HPE Private Cloud AI is built upon a cloud-native architecture using Kubernetes to orchestrate AI and machine learning workloads. To manage specialized hardware like GPUs efficiently, Kubernetes employs the device plugin framework. This framework allows hardware vendors (such as NVIDIA) to advertise their resources (e.g., GPUs) to the kubelet on a node. The Kubernetes scheduler can then account for these resources, enabling it to place pods that request GPUs onto the appropriate nodes, ensuring efficient allocation and utilization for demanding AI tasks.

B. Kubernetes ConfigMaps: ConfigMaps are used to store and manage non-sensitive configuration data for applications, not for discovering or scheduling hardware resources like GPUs.

C. Kubernetes DaemonSets: A DaemonSet ensures a pod runs on all or specific nodes. While a GPU device plugin might be deployed using a DaemonSet, the DaemonSet itself is a deployment controller, not the feature that manages the GPU resource.

D. Pod autoscaling only: Pod autoscaling (e.g., HPA) adjusts the number of pod replicas based on metrics. It does not manage the underlying hardware discovery, advertisement, or allocation of GPUs to pods.

1. Kubernetes Documentation

"Device Plugins": "Kubernetes provides a device plugin framework to enable Pods to access specialized hardware features such as GPUs... Without a device plugin

the Kubernetes scheduler is unaware of the device-specific resources on a Node

and is unable to schedule Pods that require those resources." (Source: Kubernetes.io official documentation

kubernetes.io/docs/concepts/extend-kubernetes/compute-storage-net/device-plugins/)

2. Kubernetes Documentation

"Schedule GPUs": "Kubernetes includes stable support for managing NVIDIA GPUs across different nodes in a cluster... you can schedule Pods that request nvidia.com/gpu resources. To enable this

you must have the appropriate NVIDIA drivers installed on the nodes and you must run the corresponding Kubernetes device plugin from NVIDIA." (Source: Kubernetes.io official documentation

kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/)

3. HPE Private Cloud AI Solution Brief: "HPE Private Cloud AI is a turnkey solution... built on a modern

cloud-native architecture [Kubernetes]... It provides self-service access to shared resources

including powerful NVIDIA GPUs

for AI model development and tuning." This establishes the context that the solution uses Kubernetes and manages GPUs

for which the standard mechanism is the device plugin. (Source: HPE Document a00138329enw

Page 2

"Solution overview")