- Distinguish roles: .vmcx stores configuration, and .vmrs stores execution status; they should not be deleted manually.
- Checkpoints are used for rapid reversal; they are not a substitute for independent backups.
- Updating the configuration version enables key features (vTPM, VBS, GPU‑P) on a per-host basis.
In Hyper-V, there are two files that always cause problems: .vmcx and .vmrs . If you've ever found your disk full and considered manually deleting them with PowerShell , hold on a second. Understanding what each file stores and when they're generated is key to avoiding messing things up and losing machines or leaving the host in a complete mess.
In addition to the issue of space, these files are also related to concepts such as checkpoints (snapshots) , backups, and the VM configuration version. And yes, there are also important differences between snapshots and backups, how to consolidate .avhdx disks, and when it's advisable to upgrade the configuration version to unlock new Hyper-V features.
What are .vmcx and .vmrs in Hyper‑V?
On modern Microsoft hosts, the .vmcx file stores the virtual machine's binary configuration : virtual hardware , network settings, disk paths, etc. It replaces the older XML configuration from previous versions. Without this file, the VM literally doesn't know who it is or how to boot.
The .vmrs file stores the VM's runtime state . In current versions, it coexists with .vmgs (guest state) starting with VMs version 8.2 and higher. The .vmrs file is used for saved states and certain runtime metadata; it is not a data disk, so deleting it can corrupt the VM or prevent it from resuming.
Default locations on current Windows hosts: C:\ProgramData\Microsoft\Windows\Hyper-V\Virtual Machines for .vmcx , .vmrs , and .vmgs files , and C:\ProgramData\Microsoft\Windows\Hyper-V\Virtual Hard Disks for .vhd / .vhdx disks . Checkpoints generate their own .vmcx and .vmrs files in C:\ProgramData\Microsoft\Windows\Snapshots , in addition to the .avhdx differentiation disks.
Can I delete .vmcx or .vmrs to save space?
The short answer is: don't . The .vmcx file contains the VM's identity and configuration; deleting it renders the machine unusable until you rebuild that exact configuration. The .vmrs file stores the running/saved state; deleting it will cause you to lose that state and could lead to inconsistencies.
If space is the problem, address the usual causes of growth: forgotten checkpoints that generate chains of .avhdx files, unnecessary saved states, and VMs with a lot of RAM that reserve more state. The surest way to reclaim space is to delete checkpoints from Hyper-V Manager and enable automatic consolidation of .avhdx files with the parent .vhdx file.
To prevent Hyper-V from saving state when shutting down the host (and thus creating large .vmr files), check the automatic shutdown action for each VM: set it to Shut Down or Shut Down Guest instead of Save State . This way, Hyper-V won't have to persist the runtime state when you shut down the host, and you'll avoid creating large .vmr files when they're not needed.
Snapshots (checkpoints) vs. backups: they are not the same thing
VM snapshots and backups both serve recovery purposes, but they have different functions . A snapshot captures the current state (disks, memory, devices) and then transfers changes to delta disks for that point in time. A backup is a separate copy (ideally external) designed for long-term retention and disaster recovery.
Snapshots are fast and very useful for testing or short-term, risky changes. Backups (ideally image-based) protect entire systems and can be stored on external storage or in the cloud, with deduplication, compression, and retention policies. Using snapshots as backups is a mistake : snapshots depend on the base disk, consume storage over time , and negatively impact performance if chained together or retained for days.
| Snapshot | Backup | |
| Our Purpose | Reverse quickly to the punctual state of the VM | Integral protection and full or granular restoration |
| Dependence | Depends on source data (base disks) | Copy independent in another location |
| Retention | Short (hours/days); penalizes performance if extended | Long; it is optimized with dedupe/compression |
| Recovery | Quick on the Original VM | Flexible; allows new VMs on another host |
How snapshots work and what files they create
In terms of storage, snapshots can be based on copy-on-write (CoW) , where changes are written to a delta when the original block is modified, or on redirect-on-write (RoW) , redirecting writes to a new location while leaving the original intact. In both cases, the chains are parent-dependent; if the base disk fails, there is no going back.
In VMware vSphere, a snapshot generates several files: .vmdk (descriptor and -flat with data), -delta.vmdk (SPARSE differentials), .vmsd (snapshot metadata), and .vmsn (memory state if captured). All reside alongside the main disks and grow as the VM writes more data while the snapshot is open.
In Hyper-V, checkpoints create: checkpoint-specific .vmcx and .vmrs files, the .avhdx (difference disk), and trace files such as .rct (Resilient Change Tracking, crucial for incremental backups) and .mrt (host failure resilience). In VMs 8.2+, .vmgs files for guest state also appear .
Best practices with snapshots in production
Use checkpoints as a temporary safety net before updates, configuration changes, or testing. Avoid long chains: 2-3 snapshots maximum, and per VM, do not extend beyond 72 hours to minimize the impact on I/O.
After verifying that everything is working correctly, delete the snapshots to consolidate the .avhdx files with the primary disk. If you see storage growth or performance drops, check the VM folder for chained .avhdx files and manage the consolidation from Hyper-V Manager; manual operations on unplanned delta disks can cause data loss.
VM Backups: Image vs. File and How Snapshots Work
Image-based backups capture the entire VM (OS, configuration, disks) and allow for full restores or instant boot from backup. File-based backups install an agent in the guest and protect data at the file system level, but do not restore the virtualized environment.
Modern solutions orchestrate snapshots of the platform (VMware/Hyper-V) or storage to freeze the state, copy only changed blocks (supported by CBT/RCT), and delete the snapshot as soon as the backup is consistent. This reduces the impact on production and speeds up the backup window.
There are commercial tools that leverage these techniques for ESXi and Hyper-V with features such as incremental backups , deduplication, archiving to S3, fast restore, centralized multi-host management, and automatic retention. The key is that the resulting backup is independent of the source disks and stored in separate locations (NAS, cloud, tape) to eliminate the single point of failure.
VM Configuration Versions in Hyper-V: What They Are and When to Upgrade
The configuration version defines the file format (including .vmcx/.vmrs) and the capabilities of the VM in relation to the host. When moving/importing a VM to newer hosts (Windows 10/11, Windows Server 2016/2019/2022/2025), the version does not automatically upgrade, so you can continue moving it to older hosts but you won't take advantage of new features until you upgrade.
Typical steps to review and upgrade: open PowerShell as administrator and run Get-VM To view versions, shut down the VM; in Hyper‑V Manager, Action > Update Configuration Version, or use Update-VMVersion <NombreVM>. First, make sure you won't have to return to a host with lower version.
To see what your host supports, use Get-VMHostSupportedVersionIf you need to create a backward-compatible VM, you can set the version when you create it: New-VM -Name "WindowsCV5" -Version 5.0. The default value depends on the system: Windows Server 2025 and Windows 11 24H2 accepts configuration versions up to 12.0 (and backwards compatible to varying degrees), while older releases remain at 10.x, 9.x, etc.
Features that require a minimum configuration version
Some Hyper-V features require a minimum system level. If you don't meet this requirement, they will either not appear or will not function. These include: GPU Partitioning (12.0), ARM64 Guest support (11.0), Dynamic Processor Compatibility Mode (10.0), Nested Virtualization on AMD (9.3), vNUMA/AMD improvements (9.2/9.1), VBS and Guest with Hibernation (9.0), increasing virtual devices per type to 64 (8.3), and vTPM or hot-add/hot-remove memory (7.0/6.2). Updating your configuration version unlocks these capabilities.
Types of VM files and where they are stored
| Type | Description and extension | Default route |
|---|---|---|
| Configuration | .vmcx, binary with the VM configuration | C:\ProgramData\Microsoft\Windows\Hyper‑V\Virtual Machines |
| Runtime state | .vmrs y .vmgs, running status and guest | C:\ProgramData\Microsoft\Windows\Hyper‑V\Virtual Machines |
| Virtual disks | .vhd/.vhdx, VM data | C:\ProgramData\Microsoft\Windows\Hyper‑V\Virtual Hard Disks |
| Differentiation discs | .avhdx, are created with checkpoints | C:\ProgramData\Microsoft\Windows\Hyper‑V\Virtual Hard Disks |
| Checkpoint | .vmcx/.vmrs checkpoint-specific | C:\ProgramData\Microsoft\Windows\Snapshots |
Context and key improvements since Windows Server 2016
The arrival of 2016 changed many parts under the hood of Hyper-V: the .vmcx/.vmrs formats replaced the old ones, improving reliability and configuration read/write performance, and reducing risks from storage failures.
In storage, centralized QoS appeared with Scale-Out File Server, capable of applying minimum/maximum IOPS to VMs or groups and monitoring traffic in real time. Storage Spaces Direct (S2D) also debuted for hyperconverged nodes with SAS/SATA/SSD/NVMe (SATA behind SAS HBA), 2/3-way mirroring, and aggregated or disaggregated deployment.
The network stack adopted VXLAN for SDN, incorporated Network Controller, SLB, NFV, and RAS gateway; RDMA was simplified with Switch Embedded Teaming (SET) for NIC sharing in storage traffic and Live Migration. All of this with an eye toward convergence with Azure.
The cluster upgrade process introduced a functional level that allows mixed mode while migrating hosts, followed by an Update-ClusterFunctionalLevel to unlock new features. During mixed mode, older formats are maintained; upon completion, you can upgrade the VM (requires shutdown) to the new .vmcx/.vmrs formats.
For data copying and consistency, Hyper-V transitioned from relying on external VSS to exposing Resilient Change Tracking (RCT) , facilitating efficient incremental deployments. Production checkpoints use VSS within the VM to ensure the guest is aware of the changes and performs a clean boot, mitigating risks in distributed services such as Active Directory or Exchange.
Incident resilience has improved: in the event of brief network outages, the host enters isolation mode before triggering failover; in the event of storage loss, VMs freeze and resume when the backend returns. Furthermore, automatic VM load balancing between hosts has been implemented without relying on VMM.
In terms of security, shielded VMs combine UEFI Secure Boot, vTPM, BitLocker/dm-crypt and the Host Guardian Service with attestation (AD or TPM 2.0) to prevent host administrators from accessing protected VM disks or memory; Live Migration traffic is encrypted and sensitive telemetry is restricted.
Containers (Windows and Hyper-V) with Docker management and nested virtualization were added , and ReFS became the recommended system for VMs thanks to its speed in creating fixed disks and migrating checkpoints. Other perks included: hot-adding/removing NICs, changing static memory on the fly, backing up and resizing shared VHDXs live, and a Hyper-V Manager capable of managing 2012/2012 R2 hosts with alternative credentials.
Impact of snapshots and backups on performance and storage
A snapshot retained for days generates disk delta growth and read/write layers that penalize IOPS. The more the VM writes, the larger the -delta (VMware) or .avhdx (Hyper-V) file will grow. Backups, on the other hand, are designed to optimize storage with compression and deduplication , replicate to multiple destinations, and validate integrity.
The golden rule: snapshots for quick rollback after a risky operation; backups for real protection in the medium and long term. In production, the sensible approach is to combine both: short-term checkpoints + scheduled image-based backups.
How to check versions and plan VM upgrades
Check versions with: Get-VM * | Format-Table Name, Version. If you decide to upload, shut down the VM and run Update-VMVersion <NombreVM>. If the option does not appear in the GUI, you are already on the highest version of the host.
For hosts with long-term maintenance (Server 2016, 2019, 2022, 2025; Windows 10/11 LTSC) and semi-annual channel support, the supported versions differ . Windows Server 2025 and Windows 11 24H2 support up to version 12.0, 22H2/23H2 support up to version 11.0, Server 2022 supports version 10.0, and so on. Limit the version to the lowest common value if you plan to move VMs between heterogeneous hosts.
Security and continuity: beyond hyperconverged
Data protection remains a priority. Backups and disaster recovery are essential against cyber threats (ransomware, for example). It's advisable to stay up-to-date with CERT guides and alerts ; you can consult, for reference, this public report from the CCN-CERT: ransomware document.
If you've made it this far, you already understand that .vmcx files represent configuration and .vmrs files represent state, that they can't be deleted arbitrarily, and that checkpoints don't replace backups. With a prudent snapshot policy, image-based backups using RCT/CBT, up-to-date configuration versions, and Hyper-V enhancements (QoS, S2D, SDN, production checkpoints, shielded VMs), you can keep your VMs secure, agile, and storage under control without any surprises.
Passionate writer about the world of bytes and technology in general. I love sharing my knowledge through writing, and that's what I'll do on this blog, show you all the most interesting things about gadgets, software, hardware, tech trends, and more. My goal is to help you navigate the digital world in a simple and entertaining way.