Unfortunately, you cannot create a compute cluster with a single vGPU that can scale its vCPUs independently in Azure ML. The available VM sizes in Azure are predefined with a fixed ratio of vCPUs, memory, and GPU resources. However, you can create a compute cluster with different VM sizes that can scale within a specified range of nodes, depending on your requirements
https://docs.microsoft.com/en-us/azure/machine-learning/how-to-manage-workspace#create-a-workspace