From the performance perspective it’s recommended to stay within one physical NUMA node while allocating memory to facilitate local memory access and reduce cross-NUMA resource allocation, leading to performance penalties. As both physical server models used in VMware Cloud on AWS have relatively high memory density it should not be a problem. If you are allocating more than 256/384 GB on i3/i3en instances respectively, ensure to align the vNUMA configuration to accommodate for the extra amount of memory.
Special attention should be given to scenarios where you assign to a VM more memory that is available on a single pNUMA of the host (>256/384 GiB respectively i3/i3en), however the total amount of vCPUs assigned to a VM is less than is available on a pNUMA (<18/24 vCPUs respectively to i3/i3en). For example, if you have a VM with the following resource allocation: 16 vCPUs / 512 GiB RAM running on an i3en host, you would require to manually adjust the vNUMA configuration by forcing the VM to expose two vNUMAs (amount of memory assigned to a VM is not available on a single pNUMA of the host). To achieve this, you would need to use the VM advanced setting numa.vcpu.maxPerVirtualNode = 8.