Azure VM Size Finder

Start from the minimum vCPU and memory the workload needs, then pick the workload type and region. You can also describe it in plain English, for example "16 cores, 128 GB, SQL Server, West Europe", and the finder fills in the requirements. It lists every size that meets all of them, cheapest first, priced in your region or in each size's cheapest region.

Choose the AI / ML workload: it lists only NVIDIA GPU sizes (T4, A10, A100, H100, H200, RTX PRO 6000 and others), where CUDA, PyTorch and vLLM run. Small sizes such as Standard_NV6ads_A10_v5 get a slice of one GPU (1/6 of an A10, 4 GiB of GPU memory), enough for light inference. AMD Radeon sizes (V710, V620) are built for virtual desktops, graphics and cloud gaming and do not run CUDA.

Burstable sizes earn CPU credits while idle and slow down to a baseline when the credits run out, so they only appear when you choose the Burstable workload. They suit dev/test machines, small web sites and jump boxes, not steady production load.

They are Microsoft's public USD pay-as-you-go list prices from the Azure Retail Prices API, refreshed every Monday. Your bill can be lower with an enterprise agreement, reservations, savings plans or Azure Hybrid Benefit, and it also includes disks, bandwidth and other services that are not priced here.

No. Latency to your users, data residency rules and where the rest of your resources run usually matter more than a few dollars a month. Pick your region to compare sizes like for like, and use the size pages to see the price in every region.