


Both these VM sizes have an ephemeral disk at /dev/sda that is big enough and sits completely unused. I've also tried some other VM sizes, without any luck. Use a VM size with larger cache or disable ephemeral OS. Message: The virtual machine size Standard_NC6s_v2 has a cache size of 0 bytes, but the OS disk requires 137438953472 bytes. (VMCacheSizeTooSmall) The virtual machine size Standard_NC6s_v2 has a cache size of 0 bytes, but the OS disk requires 137438953472 bytes. With non-ephemeral OS, the diff still writes back to Azure storage persistently. With ephemeral OS, the diff disk is exclusively local to the VM and will not persist data after e.g.

# az aks nodepool add -resource-group $PROJECT_NAME -cluster-name $PROJECT_NAME -name gpunp -node-vm-size Standard_NC6s_v2 -node-taints sku=gpu:NoSchedule -node-osdisk-type Ephemeral -aks-custom-headers UseGPUDedicatedVHD=true -enable-cluster-autoscaler -node-count 1 -min-count 0 -max-count 2 When you provision the VM, you also need a diff disk to write changes made during runtime. Message: The Virtual Machine size Standard_NC6 does not support Ephemeral OS disk. (VMSizeDoesNotSupportEphemeralOS) The Virtual Machine size Standard_NC6 does not support Ephemeral OS disk. The behavior of this command has been altered by the following extension: aks-preview # az aks nodepool add -resource-group $PROJECT_NAME -cluster-name $PROJECT_NAME -name gpunp -node-vm-size Standard_NC6 -node-taints sku=gpu:NoSchedule -node-osdisk-type Ephemeral -aks-custom-headers UseGPUDedicatedVHD=true -enable-cluster-autoscaler -node-count 1 -min-count 0 -max-count 2
