The following figure depicts the dual VDI and AI architecture of the solution. The vSAN Ready Node contains two NVIDIA L40 GPUs. To enable running dual AI and VDI workloads, one vGPU is dedicated to hosting the AI workload and the other GPU to the VDI workload.
One entire L40 GPU is assigned exclusively to a single virtual machine that is dedicated to running the AI workload. This is done by assigning a vGPU profile to the AI VM that uses all 48 GB of framebuffer available on an L40 GPU. The other L40 GPU is assigned to the VDI VMs. To ensure a positive end user experience, each VDI VM is assigned a vGPU profile with 2 GB of framebuffer, allowing a maximum of 24 VDI VMs to be hosted.
The following figure depicts the architecture of the validated solution, including the network, compute and graphics, management, and storage layers. This architecture aligns with the VMware Horizon pod and block design. A pod is divided into multiple blocks. Each block is made up of one or more vSphere clusters and a vCenter Server.
