Skip to main content
Choosing the right virtual machine (VM) size in Azure directly affects performance, scalability, and cost. Azure organizes VM SKUs into families tailored to common workload patterns — learning the families and a few representative sizes is far more effective than memorizing every SKU. This article overviews the core VM families, provides practical selection guidance, and points to the tools and docs you should use when picking a size.

​
Core VM families and when to use them

  • General purpose
    • Balanced CPU-to-memory ratio for a broad range of workloads such as development/test environments and small to medium databases. Representative families: B-series (burstable) and D-series (general purpose).
  • Compute optimized
    • Higher CPU-to-memory ratio for compute-bound workloads like application servers and batch processing. Representative family: F-series.
  • Memory optimized
    • Larger memory allocations for in-memory databases and analytics. Representative family: E-series.
  • Storage optimized
    • Tuned for high disk throughput and IOPS — ideal for big data, large databases, and workloads requiring heavy local storage. Representative family: Lsv2-series.
  • GPU
    • Equipped with GPUs for graphics rendering, visualization, and machine learning training/inference. Representative family: N-series.
  • High-performance compute (HPC)
    • Very high CPU performance and low-latency network throughput for tightly coupled compute jobs and HPC scenarios (e.g., H-series).
  • Confidential compute
    • Hardware-based trusted execution environments to protect data in use for sensitive workloads.

​
Quick reference table

​
How to pick the right size

Choosing a VM size is more than matching raw specs. Follow this practical checklist:
  • Identify workload characteristics:
    • CPU-bound: prefer compute-optimized sizes.
    • Memory-bound: choose memory-optimized sizes.
    • I/O-bound: select storage-optimized sizes and provision appropriate disk types (Premium SSD, Ultra Disk).
    • GPU/accelerator needs: pick GPU series and confirm driver/runtime support.
  • Select a scaling strategy:
    • Scale up (vertical) — larger VM size for single-instance performance.
    • Scale out (horizontal) — more instances with smaller sizes behind a load balancer or VM Scale Set.
  • Consider storage and networking:
    • Do you need ephemeral OS disks, accelerated networking, or specific disk throughput?
    • Ensure selected VM supports the disk types and maximum IOPS required.
  • Optimize cost and billing:
    • Compare on-demand vs. reserved instances vs. spot VMs.
    • Use tags and budgets to track costs.
  • Confirm region availability and quotas:
    • Not every SKU is available in every region.
    • Check subscription quotas before provisioning.

​
Tools and documentation to validate choices

Use the Azure portal size selector and the CLI/PowerShell commands to preview available sizes in a region before provisioning.
A presentation slide titled "Virtual Machine Sizing" showing a table of VM types (Storage optimized, GPU, HPC, Confidential computing), example sizes, and their targeted workloads. The right column describes use cases like high disk throughput, GPU rendering/ML, high-performance computing, and confidential data processing.
Choosing the correct VM size means aligning Azure SKUs with your application’s CPU, memory, storage, networking, and GPU requirements — and validating availability and cost in your target region before provisioning.
Always verify VM availability and pricing in the target Azure region. Use the Azure VM sizes documentation, the Resize VM guide, and the Azure Pricing Calculator to validate performance and cost before provisioning.

Watch Video