Hello Ramapriya Rajamani (ramrajam),
The 50% throughput degradation you are seeing after moving from Windows Server 2019 to 2025 with Hyper‑V, even with SR‑IOV enabled, is not a generic configuration issue but points to changes in the network stack and driver interaction with the new kernel. SR‑IOV itself only bypasses the virtual switch datapath; if throughput is still halved, the bottleneck is either in the NIC firmware/driver compatibility with the 2025 kernel, or in the way the vSwitch and VMQ/VMMQ queues are being handled post‑upgrade.
The first thing to validate is that the NIC drivers are the latest vendor‑certified versions for Windows Server 2025. Inbox drivers provided by Microsoft often lag behind vendor releases, and with SR‑IOV the NIC firmware and driver must match the OS kernel capabilities. For Intel adapters, confirm you are running the latest PROSet package; for Mellanox or Broadcom, check the vendor’s support matrix for Server 2025. Outdated drivers are the most common cause of degraded throughput after major OS upgrades.
Next, check whether VMQ and VMMQ are enabled and properly balanced. In Server 2019, VMQ scaling was stable, but in Server 2025 the default queue allocation logic changed. Run Get-NetAdapterVmq and Get-NetAdapterSriov on the host to confirm queues are active and mapped correctly. If you see queues pinned to a small subset of cores, you may need to adjust RSS/VMQ processor affinity using Set-NetAdapterVmq to spread load across NUMA nodes.
Also confirm that the virtual switch is configured with the correct offload settings. In Server 2025, some offloads (such as RSC and RDMA) are disabled by default in certain configurations. Run Get-NetAdapterAdvancedProperty and check that Large Send Offload (LSO), Receive Side Scaling (RSS), and SR‑IOV are all enabled.
If you are still seeing degradation after driver and queue validation, disable VBS (Virtualization Based Security) temporarily on the host. The 2025 kernel enforces additional isolation that can impact NIC performance in SR‑IOV scenarios. This is documented in Microsoft’s performance tuning guidelines for Hyper‑V.
Finally, check whether the throughput drop is consistent across all NICs and VMs or isolated to specific adapters. If it is consistent, this is likely a regression in the 2025 networking stack. In that case, you should open a support case with Microsoft and reference the performance delta between 2019 and 2025 with SR‑IOV enabled. Microsoft has historically released hotfixes for similar regressions in Hyper‑V networking after major version upgrades.
In short, update NIC drivers and firmware to 2025‑certified versions, validate VMQ/VMMQ queue distribution, confirm offload settings, and test with VBS disabled. If degradation persists, escalate to Microsoft Support as this is likely a kernel‑level regression.
I hope you've found something useful here. If it helps you get more insight into the issue, it's appreciated to accept the answer. Should you have more questions, feel free to leave a message. Have a nice day!
Domic Vo.