What Size VMs Should I Use for AVD?
The Smart Scaler Team
VM sizing is one of the most impactful decisions in your AVD deployment. Too small and users suffer. Too large and you waste money.
Quick Reference
| Workload | Recommended Series | Example SKU | Users per VM |
|---|---|---|---|
| Light (basic Office, web) | D4s_v5 | 4 vCPU, 16GB | 4-6 |
| Medium (full Office, light apps) | D8s_v5 | 8 vCPU, 32GB | 6-10 |
| Heavy (CAD, development) | D16s_v5 | 16 vCPU, 64GB | 2-4 |
| GPU (3D, video) | NV-series | Varies | 1-4 |
The General Rules
vCPU
- 2 vCPU per user is the safe starting point
- Light workloads can stretch to 1.5 vCPU
- Heavy workloads may need 4+ vCPU
Memory
- 4GB per user minimum
- Heavy workloads: 8GB+
- Don’t forget OS overhead (~2GB)
Common Mistakes
1. Oversizing “Just in Case”
Paying for D16s when D8s would work. Measure first, then size.
2. Undersizing to Save Money
False economy. Poor performance = support calls + unhappy users.
3. One Size Fits All
Different user groups have different needs. Use multiple host pools.
4. Ignoring Storage
CPU and RAM matter, but slow disks kill performance. Premium SSD is usually worth it.
How to Right-Size
- Start with Microsoft’s recommendations
- Monitor actual usage (CPU, RAM, disk) for 2-4 weeks
- Adjust based on data
- Repeat quarterly — workloads change
The Scaling Connection
Right-sizing and scaling work together:
- Right-sizing: Optimal cost per running hour
- Scaling: Fewer running hours
A 20% smaller VM + 40% fewer hours = massive savings.
Quick Tip
If you’re unsure, start with D8s_v5 (or D4as_v5 for AMD). It’s the sweet spot for most office workloads.
The Smart Scaler can help you understand your actual resource usage patterns — including whether your VMs are appropriately sized.