Recently in the 2025 Windows Server Summit, Microsoft detailed new and enhanced capabilities around Windows Server 2025 Hyper-V, including GPU partitioning, live migration enhancements and other features. With the uncertainty with VMware by Broadcom, many organizations are perhaps looking at Hyper-V once again. Let’s take a look at these new features and enhancements and see what they bring to the table.
Hyper-V Architecture footprint has been reduced & more capabilities
One of the enhancements with Windows Server 2025 Hyper-V is the reduced footprint of the Hyper-V hypervisor layer. This layer is now even more efficient and contains new features to note and expanded capabilities.
At its core, the Hyper-V hypervisor contains the following features and enhancements:
- CPU and memory isolation
- Device ownership arbitration
- Synthetic device emulation
- Partition messaging through the VM bus
Now, every virtual machine is backed by a vmwp.exe process and the host manages this executable. Going beyond vmwp.exe there is the Virtual Machine Management Service (VMMS) and WMI-based APIs that help with controlling the VM state and PowerShell interaction.
The Hyper-V architecture is now a more layered and modular approach that makes the hypervisor even more capable with things like GPU partitioning without adding any layer of complexity or unnecessary code.
Synthetic vs. SR-IOV in Hyper-V and GPU Partitioning
In previous Hyper-V releases, Hyper-V used synthetic devices for I/O operations. These synthetic drivers are the mechanism used to abstract hardware devices for guest operating systems. These can then communicate with the VM bus to the Hyper-V host.
Traditionally, Hyper-V used synthetic devices for most I/O operations. These synthetic drivers abstract hardware devices for guest operating systems and communicate via the VM bus to the host using Single Root I/O Virtualization (SR-IOV).
This technology used to be an edge use case with NICs to help reduce latency and CPU overhead by physically accessing the hardware devices. However, Microsoft has implemented SR-IOV as more of a “normal” or “standard” approach to handling hardware.
This is also a central feature with the new GPU partitioning (GPU-P) feature that allows a physical GPU to be logically separated and shared between various virtual resources.
Here’s how GPU partitioning:
- Physical GPUs can now be divided up by Hyper-V into multiple isolated virtual functions (VFs)
- Each VF can be assigned to a VM as if it were a dedicated GPU
- With this approach, multiple VMs can share the same physical GPU at the same time as maintaining high performance and driver compatibility
- Also important, now these GPU-backed VMs can now live migrate across hosts
AI, ML, CAD, and rendering workloads are becoming more popular than ever so this is a centrally important feature with Hyper-V moving forward. Live migration used to be difficult or impossible for GPU-bound VMs. Windows Server 2025 has changed this entirely and eliminated this as a blocker for drastically increased mobility.
Learn more about GPU partitioning here: Introducing GPU Innovations with Windows Server 2025 | Microsoft Community Hub.
Hardware Requirements for GPU-P and SR-IOV
Speaking of GPU Partitioning and SR-IOV, what are the requirements? There are several things that need to be in place before you can take advantage of GPU-P. These include the following:
- Compatible GPUs that support GPU-P and SR-IOV (NVIDIA and AMD enterprise-grade cards)
- Processors with change tracking support, such as AMD EPYC Milan or Intel Sapphire Rapids
- Motherboard BIOS support for SR-IOV
- Updated drivers and firmware
Microsoft provides a list of compatible hardware and recommends working closely with your hardware vendor to ensure support.
Live Migration is better in the latest Hyper-V
There are several key steps involved during a normal Live Migration task carried out by Hyper-V. In review, that process includes the following steps:
- The source host initiates a TCP connection with the destination host
- VM configuration is sent to the target host to create a “skeleton” VM
- Memory Copy Begins: The VM’s memory is copied from the source to the destination while the VM remains running
- Hyper-V performs multiple passes, copying only modified memory pages each time to reduce data size
- The remaining memory pages, CPU state, and device state are copied in a brief pause (brownout)
- The storage handle is moved to the new host
- The VM resumes running on the destination host with minimal downtime
Live migration in itself is not new. However, Microsoft has introduced some really great new improvements in Windows Server 2025 related to Live Migration. Note the following.
Live Migration compression
Data transferred during migration is compressed. This reduces the network bandwidth and speeds up migration by 2X in most cases.
RDMA-Based Live Migration
When using Remote Direct Memory Access (RDMA) NICs, Virtual machines can be migrated without tagging the CPU with the extra overhead. This is ideal for large VMs or clusters with heavy traffic and speeds up migrations without burdening the CPU.
Live Migration Network Selection Enhancements
Failover clusters used a static list of preferred networks that was configured as an ordered list. In Windows Server 2025, the cluster has intelligence about which networks it uses. This is the process that it follows:
- The cluster looks at all possible network paths using its NetFT component
- The system looks at which routes are valid and picks the best one even when networks span different subnets or data centers
- This eliminates delays from misordered static lists and supports cross-site live migration
Admins can still reorder or exclude specific networks in the Failover Cluster Manager UI, but the smart defaults now better reflect complex networking environments, including direct-connected nodes and stretched clusters.
Storage Live Migration and Shared Nothing Migration
One of the great new features related to Live Migration is the Storage Live Migration and Shared Nothing Migration. Both of these areas have been areas where I think VMware has been ahead of Microsoft for quite some time. However, with the Hyper-V release in Windows Server 2025, Microsoft is definitely stepping up their game with improved VM mobility in these areas. Let’s note the enhancements.
Storage Live Migration
Now, with Hyper-V, VMs can be moved between one storage volume and another with zero downtime. When you begin a storage Live Migration, Disk writes are mirrored between source and destination volumes. One the migration is complete, the VM seamlessly switches to the new storage location.
Shared Nothing Live Migration
Also, Shared Nothing Live Migration in Windows Server 2025 allows VMs to be transferred between Hyper-V hosts that have no shared resources. This is a great feature for moving workloads across standalone server or clusters that don’t have shared storage locations in common.
High Availability and Clustering features are still going strong
Microsoft touts the HA and clustering features in Windows Server 2025 as still going strong. Of course, these are longstanding features that have been around for quite some time in Windows Server Hyper-V and continue to get the polish of each new Windows version in terms of Failover Clustering and Hyper-V combined.
The following are the latest configuration maximums for Hyper-V clusters:
- Up to 64 nodes and 8,000 VMs per cluster with shared storage.
- Up to 16 nodes in HCI clusters using local storage.
- Automatic failover to another node if a physical host fails.
- Affinity and Anti-Affinity rules for intelligent VM placement across nodes.
With the meteoric rise of AI workloads, you can combine GPU-P with HA for your critical AI or CAD systems so these will failover without losing GPU access in the event you have a cluster node failure. This is a major advancement compared to workloads from previous versions of Hyper-V.
Generation 2 VMs are now the standard
Microsoft has held on to the Generation 1 VM default for years now. However, that all changes with Windows Server 2025 Hyper-V. Now, Generation 2 VMs are the default generation of virtual machines when you create new VMs. Now is the time to start thinking about upgrading your generation 1 VMs to generation 2.
As a review, what are the benefits of generation 2 VMs compared to generation 1?
- With generation 2, you get up to 248 virtual processors
- You also get up to 240TB of RAM per VM
- Gen 2 offers support for TPM, UEFI boot, secure boot, and hot-add hardware
- It also has better boot performance and security features
With Hyper-V in Windows Server 2025 and Windows 11, new VMs now default to Gen 2 VM creation. It will be best practice moving forward now with Windows Server 2025, Windows 11, and beyond to stick with generation 2.
Processor Compatibility Mode
With all Hyper-V versions prior to Windows Server 2025, Hyper-V relied on what was called static processor compatibility for performing Live Migration across nodes with different CPUs. However, with Windows Server 2025, it uses what they call a dynamic processor compatibility. This new dynamic mode detects the common feature set across all nodes and adjusts the VM behavior according to that standard and shared set of capabilities.
Live migration performance and compatibility are benefited as a result.
Copilot can help with your PowerShell scripting with Hyper-V
It seems like AI is everywhere these days with generative AI on the rise in tools of all kinds. With Co-Pilot in regards to PowerShell and Hyper-V, this is no exception.
At the Windows Server Summit, Microsoft showcased how GitHub Copilot has really good integration with PowerShell scripting for Hyper-V.
Note the following examples shown:
- Creating bulk VMs with fairly complex configurations
- Modifying VM settings in batch
- Automating inventory, networking, or storage setup in Hyper-V
IT automation has always been challenging if you didn’t know a scripting language well. However, now with the availability of generative AI, prompts can be used to generate the code needed for Hyper-V automation.
Summary of benefits that make Windows Server 2025 more powerful
There are many benefits in regards to Hyper-V that make this the most powerful Windows Server operating system to date.
Hyper-V in 2025 introduces several new capabilities that place it toe to toe with VMware and other hypervisors, including the following:
- GPU Partitioning with Live Migration – GPU workloads can now be virtualized without mobility of the virtual machine being affected or limited
- Massive Scalability – You can make use of up to 4PB of RAM per host, 240TB per VM, and 248 vCPUs make it enterprise-ready for even the heaviest workloads.
- Smarter Clustering and Network Logic – You get better failover intelligence with Windows Server 2025 Hyper-V that makes managing large or cross-site clusters much easier
- Integrated AI for Automation – Copilot helps to make infrastructure-as-code and intuitive scripting available to all administrators, regardless of their scripting experience.
Wrapping up
Windows Server 2025 Hyper-V brings really great new enhancements that make running production workloads much easier on Hyper-V than ever before. Many organizations are looking to use GPUs in their virtualized workloads for AI, ML, and other use cases. Now with the GPU partitioning capabilities in 2025, this will be much easier. They have definitely closed the gap with VMware and other vendors in the virtualization space with many of these features.
