https://learn.microsoft.com/en-us/azure/virtual-machines/nva10v5-series

Some instance type, like Standard_NV6ads_A10_v5, only holds a fractional of GPU (1/6 for Standard_NV6ads_A10_v5). The GPU memory of those instance types is also fractional (4GB for Standard_NV6ads_A10_v5). But they are tagged as one whole GPU and used as if they have 24GB GPU memory, like the following example.