← Back to Blog

Move a VMware VM to a KVM VPS Without VDDK: Convert VMDK or OVA to QCOW2 Step by Step

Published · Updated · by Leo Gashi, RS Computers

VMware KVM Migration

To move a VMware VM to KVM, export it as an OVA or copy its VMDK files, convert the disk to QCOW2 with qemu-img (or virt-v2v, which also puts the right drivers in), make sure the guest can boot from VirtIO disks, and load the image onto a KVM server. None of this needs VMware's VDDK, the disk library Broadcom stopped offering for public download in August 2026. A Linux VM with a few gigabytes of data is an evening's work; Windows takes longer because its drivers must go in before the move. If you are still deciding where the VM will land, our KVM VPS plans list every size, and the table in Step 5 maps typical VMware VMs to them.

We first tested this route on Debian 13 on 7 October 2026 and re-ran the qemu-img conversion steps on 18 October 2026 with qemu-img 10.0.13, checking the input methods of virt-v2v 2.6.0 at the same time. The job in five moves:

  1. Prepare the guest on VMware: VirtIO drivers in, VMware Tools out, network details written down.
  2. Shut it down and get the disk out as OVA, OVF or plain VMDK files.
  3. Convert the disk to QCOW2 or RAW on any Linux machine and check it.
  4. Boot it once on your own computer.
  5. Send it to us, and we import it onto your RS Computers VPS or VDS.

Key facts, checked on 2026-10-18:

Broadcom pulled the public VDDK downloads: what that changes

VDDK, the Virtual Disk Development Kit, is a closed-source VMware library that lets other programs read a VM's disks straight from an ESXi host or vCenter over the network. Backup software uses it, and so do most "click and migrate" tools, because it copies disks quickly while ESXi keeps serving them. Until August 2026 it could be downloaded from Broadcom's developer site.

That download is gone, with no deprecation notice. When we opened Broadcom's VDDK developer page on 18 October 2026 it returned HTTP 404, and we found no Broadcom knowledge base article about the change. Broadcom's statement to TechTarget was short: "The VDDK is not a customer entitlement." According to It's FOSS, Red Hat's support note says it cannot host or redistribute the kit either.

In practice, a VDDK copy you already have still works within its licence. Any guide that starts with "download VDDK from Broadcom" now stops at step one. Offline methods never needed it: export the VM or copy its VMDK files, then convert them with free tools. For three to ten VMs and a weekend of downtime, that is the simpler route anyway.

Why small VMware shops are leaving anyway

Since Broadcom bought VMware on 22 November 2023, perpetual licences ended and every CPU is licensed for at least 16 cores, so a host with two 8-core CPUs pays for 32. In March 2025 distributor Arrow told partners the minimum order would rise to 72 cores per product from 10 April (The Register); Broadcom withdrew it before that date, saying it had never announced a price change and that the 16-core minimum from December 2023 stands (heise, 10 April 2025). Free ESXi 8.0 Update 3e is limited to two physical CPUs and eight vCPUs per VM, with no vCenter or support (Broadcom). In a CloudBolt survey of 302 IT decision-makers at North American companies with over 1,000 staff, published in February 2026, 86% said they were actively reducing their VMware footprint (CloudBolt sells migration tooling). Replacing a whole private cloud instead? Our article on moving from VMware to OpenStack covers that.

Rebuild, or bring the whole disk

SituationBetter routeWhy
Linux web server built from a script or Docker ComposeRebuildA clean install plus your files and a database dump leaves no VMware leftovers.
Old Linux server nobody remembers setting upWhole diskThe disk is the documentation.
Windows Server with business software and licencesWhole diskMoving the machine is easier than reinstalling and relicensing the application.
A disk far bigger than the data on itEitherThe new disk must be at least the virtual disk size, so a rebuild can mean a smaller plan.

For a rebuild, our guides to moving a website to a VPS, running databases on a VPS and Docker on a VPS cover the pieces. The rest of this article is the whole-disk route.

Step 1: prepare the VM while it still runs on VMware

On VMware the disk sits behind a VMware controller (PVSCSI or LSI Logic) and the network card is a vmxnet3. On KVM both become VirtIO devices, and a guest without a VirtIO driver at boot cannot find its own disk. That is why most failed migrations fail.

Linux guests

Linux loads its first drivers from a small file called the initramfs. Debian and Ubuntu include most storage drivers in it. Red Hat, AlmaLinux, Rocky and Fedora build it only for the current hardware, so a VM built on VMware often stops at boot with a dracut timeout.

# as root on the VMware VM (Debian or Ubuntu)
lsinitramfs /boot/initrd.img-$(uname -r) | grep -E 'virtio_(blk|scsi|net)'

You should see virtio_blk, virtio_net and virtio_scsi modules. If nothing prints, add those three names plus virtio_pci to /etc/initramfs-tools/modules and run update-initramfs -u -k all. On the Red Hat family:

# as root on the VMware VM (AlmaLinux, Rocky, RHEL, Fedora)
dracut -f --add-drivers "virtio_blk virtio_scsi virtio_net virtio_pci"
lsinitrd | grep -oE 'virtio_(blk|scsi|net)' | sort -u

In our 7 October test on AlmaLinux 9 this printed virtio_blk, virtio_net and virtio_scsi; virtio_pci is built into that kernel. Then note your network settings (ip -br addr, ip route), check that /etc/fstab uses UUID= rather than /dev/sda1 (a VirtIO disk is /dev/vda), and record BIOS or UEFI: [ -d /sys/firmware/efi ] && echo UEFI || echo BIOS. Finally swap open-vm-tools for qemu-guest-agent.

Windows guests: VirtIO drivers first

Without the VirtIO storage driver, Windows stops with INACCESSIBLE_BOOT_DEVICE (0x7B). The order that works:

  1. Inside the VM, run virtio-win-guest-tools.exe from the virtio-win 0.1.302 folder linked above. It installs the disk, network and balloon drivers plus the QEMU guest agent.
  2. Uninstall VMware Tools and restart, while still on VMware: Broadcom's knowledge base says the uninstaller fails once the VM has left vSphere.
  3. Turn off Fast Startup in Power Options. A hibernated disk is not safe to convert, and virt-v2v refuses it.
  4. Save ipconfig /all, suspend BitLocker if it is on, and shut down from the Start menu.

Step 2: get the disk out of VMware without VDDK

Where the VM livesHow to get the diskWhat you get
ESXi host clientPower off, right-click the VM, ExportAn .ovf plus one .vmdk per disk
vCenterActions, Template, Export OVF Template; "Single file" gives an OVAAn .ova, or a folder of files
ESXi with SSH onCopy the small descriptor .vmdk and its -flat.vmdk with scpA flat disk as large as the full virtual disk
Workstation or FusionCopy the VM folder's .vmdk files, or File, Export to OVFSparse or split VMDKs, or an OVA

The SSH copy looks like this. The host and datastore names are examples, and we had no ESXi host to run it on:

# as your normal user on your Linux machine (ESXi SSH service turned on)
scp root@esxi01.example.com:/vmfs/volumes/datastore1/web01/web01.vmdk .
scp root@esxi01.example.com:/vmfs/volumes/datastore1/web01/web01-flat.vmdk .

An OVA is an uncompressed tar archive holding the .ovf description, the disks and a checksum file; tar -xvf yourvm.ova unpacks it. The .ovf is readable XML with the CPU count, memory and firmware type, which helps you choose a plan.

Step 3: convert VMDK to QCOW2 with qemu-img

Use any Linux machine with room for the converted image: your PC, a Linux VM, or WSL on Windows. We used a fresh Debian 13.7 container with 2 GiB of memory. With no vSphere host for this re-test, we made the VMware disks ourselves: qemu-img wrote the 21 MB CirrOS 0.6.3 test image (MD5 checked) as a compressed streamOptimized VMDK, the kind inside OVA files, and as a monolithicFlat pair, the layout ESXi keeps on its datastore.

# as root on your own Debian 13 or Ubuntu machine
apt update
apt install qemu-utils
qemu-img --version

We saw qemu-img version 10.0.13 (Debian 1:10.0.13+ds-0+deb13u1). Look at the disk first:

# as your normal user, in the folder with the exported files
qemu-img info web01-disk1.vmdk
# what qemu-img printed in our lab (trimmed)
file format: vmdk
virtual size: 112 MiB (117440512 bytes)
disk size: 20.3 MiB
    create type: streamOptimized
            compressed: true

The virtual size is the disk the VM sees; your new plan's disk must be at least that big. The disk size is what the file really uses. Now convert, check and compare:

# as your normal user
qemu-img convert -p -f vmdk -O qcow2 web01-disk1.vmdk web01.qcow2
qemu-img check web01.qcow2
qemu-img compare -f vmdk -F qcow2 web01-disk1.vmdk web01.qcow2
# what we saw
No errors were found on the image.
354/1792 = 19.75% allocated, 0.00% fragmented, 0.00% compressed clusters
Image end offset: 23527424
Images are identical.

Only about 20% of the virtual disk holds data, so the QCOW2 file is 22.4 MiB rather than 112 MiB. qemu-img compare confirms the guest will see exactly the same bytes; run it before deleting anything.

Flat ESXi disks and RAW output

A disk copied from a datastore comes as two files. The small one is a text descriptor:

# output of: cat db01.vmdk (trimmed)
# Disk DescriptorFile
createType="monolithicFlat"
RW 229376 FLAT "db01-flat.vmdk" 0

Give qemu-img the descriptor, with the -flat.vmdk beside it. For RAW instead of QCOW2:

# as your normal user
qemu-img convert -p -f vmdk -O raw db01.vmdk db01.raw
ls -ls db01.raw

The RAW file showed 117,440,512 bytes but used only 22,044 KiB, because empty parts stay as holes. Compared with the QCOW2, it again printed "Images are identical." Split Workstation disks (db01-s001.vmdk and so on) convert the same way from the small first file; files like web01-000001.vmdk are snapshot layers, so consolidate snapshots in VMware first. Skip the -S 0 option some guides recommend: the QEMU documentation says it turns off the search for empty space, making the output fully allocated.

A smaller copy for the upload

# as your normal user
qemu-img convert -p -c -f vmdk -O qcow2 web01-disk1.vmdk web01-upload.qcow2
sha256sum web01-upload.qcow2

-c compresses the QCOW2. In our 7 October test a 1.1 GB Debian disk became a 416 MB upload; the already compressed CirrOS image only went from 23.5 to 21.7 MB. Send the checksum line with the file.

virt-v2v without VDDK: the input methods that still work

qemu-img copies the guest as it is. virt-v2v also edits it: it rebuilds the Linux initramfs with VirtIO drivers, fixes /dev/sda names in /etc/fstab, removes VMware Tools on Linux and adds the QEMU guest agent. On Debian 13, virt-v2v --machine-readable listed the inputs disk, libvirt, libvirtxml, ova and vmx. From the Debian manual:

MethodNeeds VDDK?Use it when
-i disk web01.vmdkNoYou have one VMDK. We ran this one in our 7 October test.
-i ova VM.ovaNoYou exported an OVA or OVF folder from vSphere.
-i vmx guest.vmxNoYou copied the .vmx and disks, or NFS-mounted the datastore.
-i vmx -it ssh ssh://...NoESXi has SSH on (off by default) and the guest has no snapshots.
-ic vpx://..., no transportNoYou only have vCenter access; the manual calls it the slowest.
-it vddkYesYou already hold a VDDK copy.

The latest upstream manual adds an NFC method (-it nfc) that reads from ESXi port 902 without VDDK; the 2.6.0 package in Debian 13 does not have it yet.

# as root on your own Linux machine (Debian 13), with /dev/kvm available
apt install virt-v2v guestfs-tools
mkdir -p /var/tmp/out
export LIBGUESTFS_BACKEND=direct
virt-v2v -i disk web01.vmdk -o local -os /var/tmp/out -of qcow2

On our 7 October Debian guest this took 26 seconds and printed "This guest has virtio drivers installed." The output is /var/tmp/out/web01-sda (QCOW2) plus web01.xml. For Windows, download the virtio-win ISO and run export VIRTIO_WIN=/path/to/virtio-win-0.1.302.iso first, but still install the drivers inside VMware: the virt-v2v manual warns that removing VMware Tools from Windows during conversion often fails.

Step 4: boot it once on your own machine

# as root (or a user in the kvm group), after: apt install qemu-system-x86
qemu-system-x86_64 -enable-kvm -cpu host -m 1024 -smp 2 -nographic \
  -drive file=web01.qcow2,if=virtio,format=qcow2 \
  -nic user,model=virtio-net-pci

This gives the disk the same VirtIO hardware it gets on our servers. A Linux guest ends at a login: prompt; virtio_blk virtio1: [vda] in the log means the disk driver loaded. Press Ctrl+A, then X, to stop. For Windows, drop -nographic and watch the window. A dracut timeout or a 0x7B blue screen sends you back to Step 1.

Step 5: pick the plan and city, then send us the disk

Choose a plan whose disk is at least the virtual size, with memory close to what the VM had. All plans are KVM with NVMe storage, 1 to 100 TB of traffic a month and their own IPv4 and IPv6. VPS plans share their vCPUs with other customers on the host; VDS plans have dedicated vCPUs, which suits databases and Windows servers that need steady CPU time.

Your VMware VMPlanSpecsNote
Small Linux VM, disk up to 20 GBVPS Nano1 shared vCPU, 1 GB RAM, 20 GB NVMe, 1 Gb/sLinux only
Linux web or mail VM, up to 40 GBVPS Micro2 shared vCPU, 2 GB, 40 GB, 1 Gb/sLinux only
Linux app server or light Windows Server, up to 80 GBVPS Mini4 shared vCPU, 4 GB, 80 GB, 1 Gb/sSmallest plan with Windows
Windows Server with business software, or a database, up to 240 GBVDS Small4 dedicated vCPU, 8 GB, 240 GB, 10 Gb/sSteady CPU
Busy file, ERP or database server, up to 480 GBVDS Medium8 dedicated vCPU, 16 GB, 480 GB, 10 Gb/sRoom to grow
Several old VMs on one machine, up to 960 GBVDS Large16 dedicated vCPU, 32 GB, 960 GB, 10 Gb/sLargest plan

You can upgrade later from the client area with a short reboot. To be fair about limits: a single VPS replaces a VM, not a cluster. If you rely on vMotion and automatic failover, or one disk is over 960 GB, plan a rebuild or a different design.

CityPick it whenOur measurement
AmsterdamUsers across western and central EuropeLondon 6.7 ms, Frankfurt 6.6 ms, Paris 9.3 ms
PrishtinaUsers in Kosovo and the BalkansSkopje 3 ms on our own network, Vienna 18.6 ms
DublinUsers in Ireland, or data kept thereSee the Ireland page

Figures are the lowest of about 10 ping.pe pings to our servers on 6 October 2026; your ping depends on your ISP. Availability by city is on the plans page. Windows is chosen at the configure step; see our Windows VPS page.

Then message us on Telegram or email info@rscomputers-ks.com with the operating system, BIOS or UEFI, the virtual disk size and the SHA-256 checksum. We arrange the upload and import your converted .qcow2 or .raw onto your VPS or VDS; like other setup work, it is quoted per job.

First boot on the VPS

Use the console in the client area for the first login, since the network may not be up. On Linux, the card was usually ens192 or ens33 on VMware and gets a new name such as ens3 on KVM:

# as root, in the VPS console
ip -br link
grep -rn ens192 /etc

The first command shows the new name; the second finds files still using the old one, such as /etc/network/interfaces or /etc/netplan/*.yaml. Put in the new name and the addresses from your service details, then reboot. On Windows, the old vmxnet3 adapter stays hidden with your old static IP: in Device Manager choose View, Show hidden devices, uninstall it, then set the address on the VirtIO adapter. Windows may also need reactivation, and Windows Server licences you bring must qualify under Microsoft's Flexible Virtualization Benefit (Software Assurance or a subscription; plain OEM or retail licences do not).

When the converted VM will not start

What you seeLikely causeFix
"dracut-initqueue timeout" or an emergency shellNo VirtIO drivers in the initramfsRun the dracut command from Step 1, or convert with virt-v2v.
Blue screen 0x7BVirtIO storage driver missingInstall virtio-win on VMware, then export again.
"No bootable device" or a UEFI shellBIOS and UEFI mixed upTell us the firmware type; on UEFI Linux, grub-install --removable adds the fallback boot file.
Boots, but no network or a missing diskRenamed card, or /dev/sdb is now /dev/vdbFix the card name; use UUID= entries in /etc/fstab.

Keep the original VM switched off but not deleted until the new server has run for a week. It is your rollback.

Frequently asked questions

Do I need VDDK to migrate from VMware to KVM?

No. VDDK only lets tools read disks live from ESXi or vCenter. Exporting an OVA or copying the VMDK files, then converting with qemu-img or virt-v2v, works without it. In the virt-v2v 2.6.0 manual in Debian 13, only the -it vddk method needs the library.

Can I still download VMware VDDK?

Not publicly. Broadcom's VDDK download pages stopped working in August 2026, and its developer page returned HTTP 404 on 18 October 2026. Broadcom says access continues through select Technology Alliance Program partners for backup and recovery. Copies you already have keep working.

How do I convert VMDK to QCOW2?

Install qemu-utils on Linux and run qemu-img convert -p -f vmdk -O qcow2 disk.vmdk disk.qcow2, then qemu-img check and qemu-img compare. For a flat or split VMDK, give it the small descriptor file with the other pieces in the same folder.

Should I send QCOW2 or RAW?

QCOW2 stores only used space and can be compressed with -c, so it is the better upload format. RAW is a byte-for-byte copy as large as the virtual disk. RS Computers can import either.

Does qemu-img convert change anything inside the VM?

No. It only changes the disk's container format. Drivers and network names must be handled in the guest, or by virt-v2v, which does edit it.

Can RS Computers import my converted VMware disk?

Yes. Send the converted .qcow2 or .raw with its checksum and we import it onto your KVM VPS or VDS. It is quoted per job, like other setup work, via Telegram or email.

From VMDK to a running server

Drivers in, VMware Tools out, a clean shutdown, then qemu-img convert, check and compare, a test boot and a compressed upload copy. Pick one of our VPS plans at least as big as the virtual disk, send the details on Telegram, and keep the old VM for a week.

← All articles

Chat on Telegram