Easily Scale AI/ML Workloads with VMware vSphere

Key Takeaways

managing ai and ml workloads within the enterprise data center can be time consuming and resource intensive it admins need tools that are easy to use and have the flexibility to easily scale as resource demands increase nvidia ai enterprise with vmware vsphere makes scaling of resources and managing workloads easy for both it admins and ai practitioners this example the ai practitioner is training an image classification model in a jupiter notebook on a host with one gpu they leverage a pre-trained model which is included in nvidia ai enterprise and perform transfer learning during this development phase the training job is executed on a single gpu and node resulting in an execution time of roughly 180 seconds per epoch an it administrator provisioned this environment on mainstream gpu accelerated servers running vmware vsphere with chanzu and nvidia ai enterprise to size tanzu kubernetes cluster nodes the it admin creates a virtual machine class with vcenter to right-size resource reservations on the vm for processing power the kubernetes cluster is deployed once and ai frameworks and vm workers are standardized across the enterprise reducing the burden on it to create additional environments as ai workloads scale the tensorflow container leveraging the gpu and mpi operators which are backed by nvidia enterprise support are built into a custom container ai workloads are now scaling across two gpu accelerated nodes in the tonzu kubernetes cluster doubling the training performance further improvement in training performance can be achieved by scaling to more nodes learn how to scale and manage the entire ai lifecycle with ease launchpad is a free program where it admins can get hands-on experience deploying an on-demand gpu accelerated kubernetes cluster and iet practitioners learn how to scale training using horrible apply today [Music]