Windows Server 2025 VM constantly 50%+ CPU and feels laggy on SSD storage (other VMs fine)

J

johnr2k

New Member
Jul 3, 2025
6
0
1
Hi all,

I’m hoping someone can help shed some light on a performance issue I’m seeing with a Windows Server 2025 VM.

Host Specs​

Server: HPE DL380 Gen9
RAM: 64GB DDR4
CPU: 2 x Intel Xeon E5-2680 v3 (24 cores total, 48 threads)
Storage: SSD (local storage, same SSD used by other VMs without issue)
Proxmox version: Lastest version


Currently running VMs:
  1. Debian
    • 12GB RAM
    • 4 vCores
    • 60GB disk
    • Runs perfectly fine
    • VirtIO NIC
  2. Ubuntu
    • 8GB RAM
    • 4 vCores
    • 60GB disk
    • VirtIO NIC
    • Also runs fine
  3. Windows Server 2025 (Problem VM)
    • 12GB RAM (12288 MiB)
    • 1 socket, 4 cores (4 vCPUs total)
    • Intel E1000 NIC
    • CPU type: host
    • NUMA: disabled
    • Disk: 100GB
    • Bus: VirtIO SCSI
    • Cache: Default (No cache)
    • Discard: enabled
    • IO thread: enabled
    • SSD emulation: enabled
    • Asynchronous IO: io_uring
Ballooning device is enabled, but minimum memory is also set to 12288 MiB.

The problem is that Windows Server 2025 VM constantly shows around 50–60% CPU usage in Proxmox, even when mostly idle. Memory usage is fine (around 30%).

Inside Windows:
  • Initially saw TiWorker and Windows Modules Installer Worker high CPU (expected after install). System Interrupts service is also a service that comes up on the top when sorting by highest CPU usage
  • Ran DISM cleanup.
  • Activated Windows.
  • Set power plan to High Performance.
  • Added Defender exclusions.
  • Rebooted multiple times.

Despite that, the server still feels laggy:
  • Server Manager takes a long time to open.
  • General UI responsiveness feels delayed.
  • IIS (running locally on this server) is slower than expected when accessed from another PC.
  • Clicking around sometimes feels sticky.

Other VMs on the same host and same SSD do NOT show this behaviour.

What I’ve Checked​

  • CPU type is set to host.
  • 1 socket, 4 cores.
  • NUMA disabled.
  • SSD storage.
  • Other VMs on same storage are fine.
  • RAM increased from 8GB to 12GB.
  • Host is not under memory pressure.
Host overall load does not appear excessive.

Questions​

  1. Is 50–60% CPU usage on Windows Server 2025 considered normal in a VM like this?
  2. Would using "Default (No cache)" for disk cache impact responsiveness?
  3. Should I try write-back cache instead?
  4. Any known performance quirks with Windows Server 2025 on Proxmox?
  5. Is there anything in the above config that stands out as suboptimal?
  6. Could the NIC being E1000 be a big contributor to this?

Given that other VMs (Linux however) on the same hardware run smoothly, I’m trying to determine whether this is:
  • A Windows Server 2025 behaviour issue
  • A disk cache configuration issue
  • Something subtle in CPU scheduling
  • Or something else entirely

Any guidance would be greatly appreciated.

Thank you.
 
johnr2k said:
Is 50–60% CPU usage on Windows Server 2025 considered normal in a VM like this?
Click to expand...
Absolutely not.
johnr2k said:
Is there anything in the above config that stands out as suboptimal?
Click to expand...
Yes.
johnr2k said:
CPU type is set to host.
Click to expand...
Sometimes this works ok but often not. AFAIK it triggers CPU mitigations in Windows (like for Spectre, Meltdown, or Downfall) which then slow down the VM.
So ... choose one of the more general types which resembles your real CPU as close as possible.
 
Last edited:
  • Like
Reactions: Johannes S and johnr2k
MarkusKo said:
Try to switch cpu type "host" to "x86-64-v3", this should disable VBS inside Windows and help with the performance a bit.
Click to expand...

And when HVCI is enabled, it's only natural that performance slows down because tasks meant for hardware processing are handled by software instead.

https://learn.microsoft.com/en-us/w...ed-protection-of-code-integrity?tabs=security

It's by design that CPUs from the 7th generation and earlier, which are not compatible with Windows 11, cannot overcome this limitation.

*Even though the CPU has this hardware support, it cannot currently be used with QEMU/KVM. We can expect it to be added in the future.

*Implementing Windows specifications through a hypervisor incurs costs. Using it on unsupported hardware remains strictly within the realm of testing or experimentation. Do not use Windows Server 2025 on the 2680v3 for production environments.

*The same applies to Windows Server 2025, which is based on Windows 11.

U

[SOLVED] Post in thread 'CPU Arguments cuts passthroughed NVMe performance by 4 times'

I don't feel like going into detail about eac.

We have not experienced significant performance impacts outside of hypervisor-protected code integrity (HVCI).

https://learn.microsoft.com/en-us/w...ed-protection-of-code-integrity?tabs=security

The impact in HVCI is due to KVM's lack of support for Intel MBEC/AMD GMET, which is by design.

*It will likely be added as a feature at some point.

https://lwn.net/Articles/1051782/

I use a host Type CPU and Windows 11, but as long as Memory Integrity Features are disabled, I...
  • Like

While support may eventually come, 3rd generation CPUs lack MBEC/GMET, so even if MBEC/GMET becomes available, no improvement is expected.

https://lwn.net/Articles/951240/

Additionally, there are no workarounds to mitigate the reduced I/O speed caused by CPU vulnerabilities like Spectre/Meltdown, so you will also be affected by those.

*This does not mean your issue is caused by this. However, please understand that ultimately, you are still running it on hardware that does not meet the requirements for using Windows 11.
 
Last edited:
An update on this: I changed cpu from host to x86-64-v3 and also switch NiC from Intel E1000 to VirtIO and it’s made a huge difference. Now the cpu usage is staying at around 7 to 9% on average unless doing some task or when server is starting up
 
You must log in or register to reply here.
Top Bottom
Back