General Availability: Ephemeral OS Disk support for v6 Azure VMs

Why Ephemeral OS Disks Matter

Ephemeral OS disks operate entirely on the VM’s local SSD—often NVMe—bypassing remote Azure Storage. This local placement delivers:

Ultra-low latency for OS operations
Up to 10× faster performance versus managed persistent disks
Lightning-fast reimaging, ideal for stateless, scale-out, or microservices environments

Being non-persistent means they’re perfect when OS state can be regenerated on reboot.

v6 VM Series Now Fully Supported

As of May 27, GA support is live for NVMe‑backed Ephemeral OS disks on the v6 VM series, such as:

Dads_v6
Ddsv6
Daldsv6 / Dldsv6
And other NVMe-capable SKUs

This unlocks direct NVMe access for OS, delivering the fastest possible disk I/O experience.

Getting Started

ARM Template (excerpt)

"storageProfile": {
  "osDisk": {
    "diffDiskSettings": {
      "option": "Local",
      "placement": "NVMeDisk"
    },
    "caching": "ReadOnly",
    "createOption": "FromImage"
  },
  "imageReference": {...}
}

CLI

az vm create \
  --resource-group myRG \
  --name myVM \
  --image yourImage \
  --size Standard_D2ads_v6 \
  --ephemeral-os-disk-placement NVMeDisk \
  --ephemeral-os-disk true

PowerShell

Set-AzVMOSDisk -DiffDiskSetting Local -DiffDiskPlacement NVMeDisk -Caching ReadOnly

These commands fine-tune OS disk placement on v6 series to leverage local NVMe disks (youtube.com).

Choosing the Right Placement & VM Size

NVMe Disk Placement (GA) – best latency, highest IOPS; requires v6 sizes with NVMe.
Temp Disk Placement – available if VM has a dedicated temp disk (e.g., v5).
Cache Disk Placement – older pattern for pre-v5 VMs.

Ensure your OS image size is ≤ local disk capacity (e.g., 30 GiB fits well into a 110 GiB NVMe disk on Standard_D2ads_v6) (learn.microsoft.com).

Ephemeral vs. Persistent Disks: Key Trade-offs

Feature	Persistent OS Disk	Ephemeral OS Disk (v6 NVMe)
Data durability on stop/restart	✅ Retained	❌ Deleted
Reimage speed	Slower	⚡ Secs
Disk latency	Moderate	Ultra-low
Supported operations	Snapshots, backup	Not supported

Ephemeral disks eschew data retention but excel in speed and agility.

Advanced Scenarios

Trusted Launch / VMGS:
Ephemeral disks with Trusted Launch reserve ~1 GiB for VMGS metadata; reimages thus wipe secret state.
Confidential VMs:
Similar to Trusted Launch, they reserve disk space and adopt same reimage behavior .
OS Disk Size Limits:
Max OS size is local disk capacity (NVMe/temp/cache), typically 110 GiB in v6. Exceeding this triggers errors.

So, Where Does This Shine?

Ideal workloads include:

Kubernetes/AKS nodes – stateless nodes reimaged quickly
CI/CD build agents – spin up fast, discard clean
Microservices clusters – ephemeral OS; state in managed data disks
Auto-scaling compute jobs – high churn, low persistence need

Next Steps

Review supported VM SKUs (e.g., Dads_v6, Ddsv6…)
Resize local OS disk to use full NVMe capacity
Deploy via ARM/CLI/PS with --ephemeral-os-disk settings
Test reimage speed, resilience under your workload

TL;DR

What’s new: GA support for NVMe-based Ephemeral OS disks on v6 VMs
Why it matters: Up to 10× faster performance, reimaging in seconds
Trade-offs: No OS persistence on reboot; no snapshot or backup support
Best use cases: Stateless, scalable, ephemeral compute workloads

Feel free to drop questions or feedback on the blog – or try it out and share your experience!

References

GA announcement – performance & support details (techcommunity.microsoft.com)
Learn docs – placement options, size constraints (learn.microsoft.com)

Shailesh Vaja

P.S. Modern AI tool has been used for creating some of the content. Technical validation and proofing are done by the author.