Imaging Windows 10 v2004 with UEFI - GPT Partitions onto 31GB or smaller drive. (and failing)

  • Ok, to set the stage:

    Using FOG Stable 1.5.8, I have 4 Windows 10 v2004 Images: 2x HOME 2x PRO, 1 of each Legacy / Non-UEFI and 1 of each UEFI.

    Image Details.JPG

    They were created using the Microsoft Windows 10 v2004 iso (current image as of this typing) and they image perfectly fine to larger drive computers without issue.

    The issue I have is a portable image I make for use on thumb drives with Parted Magic / Clonezilla, I’ve been making these images since Windows 10 v 1809 onto VM drives that are 28 GB in size. (So I can image onto drives that are 32GB etc on machines without an Ethernet Port or for other situations, then just use Gparted to resize the main partition)

    This actually still works with the Non-UEFI Win10 v2004 images, but not with the UEFI versions. I am fairly certain this is due to the v2004 adding a partition to both versions, Legacy up from 2 partitions to 3, and UEFI from 4 partitions to 5.

    The error I believe comes in when FOG has to re-juggle the new partition in with the mix then accidentally overlaps another partition. Since the new version just haphazardly adds a partition it is probably mucking up some FOG code that could normally handle it, but the smaller size HDD just causes it to give up, but I’m still certain <31GB is plenty space for the UEFI image, but how to make it work? Or will this require some sort of adjustment in later versions of FOG?

    FOG Error.jpg

    Now I know my situation is unique, but it still spells incompatibility with FOG and imaging win 10 UEFI v2004+ onto devices with drives =<32GB in size.

    In other news I know I can use the advanced clonezilla menu to work around this, but I’d love to hear any other suggestions for rebuilding a working UEFI image onto a drive <32GB with parted magic or other tools that will allow the same functionality.

    Thank you guys in advance!

  • Alrighty, so my work-arounds for this issue were to use a virtual drive of 28GB to make the original image, alternatively you can delete the 4th partition to make imaging completely compatible, but you lose the ability to do an OS reset in Windows.

    If you move the partition to where it used to be in v1909, Windows can’t find it and that also loses the ability to reset, so if you have to move it, just delete it i guess…

    another alternative if using clonezilla you can image and use gparted to first move the 4th partition to the end of the drive, then resize the 3rd partition, this seems to work fine without losing the option to reset, tested on 2 different machines, a Clevo Laptop and an HP All in One.

    Otherwise typically just make sure the drive you are imaging to with FOG is generally larger than the source disk you imaged from and everything should be fine… generally speaking…

  • @george1421 interesting, Your setup sounds great for support of a large company with varied users.

  • Moderator

    @ProfDrSir said in Imaging Windows 10 v2004 with UEFI - GPT Partitions onto 31GB or smaller drive. (and failing):

    which fat standard are you using?

    There are two methodologies here. The first is a thin image in that only the windows OS and updates are added to the reference image. All other applications like office, SAP, acrobat reader are added into the image on the target computer. The advantage to this is that when updated versions of acrobat reader or other applications come out you don’t need to rebuild your reference image (because it only contains windows + updates). You just change your post deployment tasks (snapins, or what every deployment tool you are using to deploy the later application)

    The second is a fat image, in that you load all applications onto your reference image before you capture it with FOG (understand if you have applications that are based on a unique guid, like enterprise antivirus you must install those post imaging anyway). All of the time of loading these applications goes into the reference image and not post deployment. With MDT it takes about 1.5 hours to create and configure the reference image. When I capture it with fog the reference image is about 98% of the completed image. With a fat client I can take a workstation and go from bare metal to ready to move to the user’s worksite in 15 minutes. The system is fully configured and ready to roll. I don’t have to do much post install configuration.

    To recap a thin install gives you the flexibility to modify the deployment with different applications or configurations at deployment time. A fat image you put all of your work in to the reference image up front and then enjoy rapid deployment on the back end.

    Before win10 our methodology was to use the thin image approach because we could create a standard OS reference image and use it for a year or so. Basically the applications would update version more often than the OS. Now with Win10 we have to build a new OS every 18 months so the OS updates more frequently than the applications. That makes the fat image approach more beneficial.

    As I mentioned a few times previously we use MDT and the lite touch approach to build our reference image. That way we can get a consistent build each time we rebuild our reference image. When we go between 1903 and 1909 we just copy our task sequences between builds and we get the same results (hopefully) when we build a new base image on the next release of windows 10 (the last OS you will ever have/need).

  • @george1421 which fat standard are you using? and what other benefits come from using it?

  • Moderator

    @ProfDrSir said in Imaging Windows 10 v2004 with UEFI - GPT Partitions onto 31GB or smaller drive. (and failing):

    we can just remove / ignore that 4th partition it makes imaging slightly more tedious, but functionally should be the same as before I’d assume.

    This is just my personal opinion but that recovery/restore partition really adds no value in a business environment. Using the windows recovery method it take an hour or more to bring a broken computer back. With FOG, and fat win image, I can go from bare metal to ready for the user in 15 minutes. Total tech hands on time is about 40 seconds for imaging vs dealing with the restore process. Now if you have irreplaceable data on the broken system your only option is to recover especially if bitlocker is involved.

    In our case we never do an in place upgrade, its always nuke and pave when going between releases. I know the upgrade process works, its just we find its faster for nuke and pave (once the user’s profiles are backed up with USMT).

  • @george1421 I’ll be excited to see your results! if we can just remove / ignore that 4th partition it makes imaging slightly more tedious, but functionally should be the same as before I’d assume.

    I also noticed that upgrading older v of Windows 10 on Dells with the new Bios sometimes breaks Windows… X-D which has required me to image the SDD in another machine before v2004 will run on the newer Dells… FYI.

  • Moderator

    @ProfDrSir I was able to complete the MDT test build of 2004 this afternoon. I had to update my base image to use office 365 and to include the new edge browser plus a few other add on application updates (we use a fat windows image for deployment). I’m going to capture and deploy it tomorrow AM to a VM and then a physical machine for validation. In MDT I told it to not create the restore partition so the 😄 drive partition is the last partition on the disk like I suggested. If I remember right my reference image disk is 60GB so I should be able to test deploying to a 30GB VM (assuming that my fat image fits in 30GB) to confirm what you are experiencing.

  • Well one experiment I’ve done with clonezilla seems to work for imaging, I restored the image onto a bigger drive, which ofc leaves all the extra unallocated space after the partitions.


    I then move the last partition, telling it to leave 1MiB after


    Then resized sda3 to take up the remaining space:


    Seems to leave a random 1MiB after sda4, but this still works sometimes and windows loads up without issue, I think it’s because the positioning of sda4 is taken seriously by Linux, but microsoft / windoze has always been lazy with how it uses it’s partitions, especially concerning location, (ofc other than the boot partition) leading me to think the positioning of this partition as George suggested might not be all that important, but I still have to wonder what the hell it’s purpose is… I have seen GParted fail when trying to move this partition to other locations though, specifically when I was trying to manually build a smaller image, the apply steps would fail when manipulating that 4th partition.

    I don’t know if this’ll help anyone or anything, but it gives me a path forward on the thumb drive image side of things anyway. Maybe I’ll prep as small a drive as I can get away with then use that and see where it gets me with FOG.

  • @sudburr Yeah it’s definitely all connected to the same issue pertaining specifically to Win10 v2004. I do wonder about the thinking behind this change… the “why”.

  • @george1421 Yeah it has been problematic, that last partition also seems to be custom sized by win10 2004 and places it purposely at the end of the drive, almost like it’s written backwards from the end, which gparted doesn’t like. I was trying to manually copy partitions to a smaller drive but gparted habitually leaves 1 MB of unused space after the last partition, which seems to ‘break’? this goofy tacked on partition.

    But I’ll test what you posted and see what happens.

    The golden image in this case was created on VMware workstation 15.5, but I’ve done a few test installations on a Clevo P157SM-A, the weird thing tho, the new windows v2004 iso will install a v1909 on a lot of older hardware. The new v2004 has been an odd duck of an update insomuch as how it installs.

    In reference I had an old AiO that wouldn’t upgrade to 2004 and had this to say about it:


    The update from an older v to 2004 also has 2x phases, I assume 1 is to place that partition on the end of the drive.

    One thing I did notice, was the VMware Virtual drive is 60GB, so like you said that partition, if attached will require it be stuck somewhere specific on the drive… meaning the 60GB was because the golden image was built on a 60GB? … so basically whatever image I author will always require that size as a minimum to be able to image it properly…

    I did make a test build on a virtual 30GB drive, but it failed to image to a 32GB SSD…

  • Looks familiar.

    See my thread.

  • Moderator

    @ProfDrSir I guess two things.

    1. What was the golden image created on?
    2. IMO that /dev/sda4 is a problem since its a non-resizable partition and may have a fixed location on the disk. Where its start sector may be beyond the end of that 30TiB disk.

    As a test delete it on the source (golden image) server then capture it with fog. This will leave the “C:” drive as the last partition on the disk. That /dev/sda4 is probably the windows recovery partition. If that work and you “need” the recovery partition then recreate the golden image disk structure to place the recovery partition before the “C:” drive partition.

  • So here’s what the Windows 10 v2004 HOME - Non-UEFI partition looks like in GParted:

    GParted W10HN.JPG

    3 partitions, images and fits snugly on a 28GB HDD.

    Then here’s the HOME - UEFI version:

    GParted W10HU.JPG

    4 partitions, but through FOG requires a 60GB+ HDD.

  • Other additional info:

    I am currently using Virtualbox to pull the FOG image onto the virtual HDD then using PartedMagic + Clonezilla to create the final image. This could be an issue with this new image + FOG + VirtualBox, but it still only affects the UEFI images of v2004.

    I will try with larger Virtual Drives and report back the results.


    So I was wrong about the UEFI partitions, it doesn’t add a partition to the UEFI, but it does move a 4th partition AFTER the main one. I do know it adds a partition to the non-EFI image tho. Here’s a screenshot from Gparted from a successful image to a 64GB Virtual Drive, I’ll keep testing and see how small the drive can get before it trips over itself:


    Notice ~54GB isn’t used of this 64GB drive, suggesting we’d only NEED 12ish, (I know win 10 wouldn’t function in such a small space, but I’m just pointing out the issue is with the imaging, not the size of the HDD itself.)

    I’ve found that this works with a virtual HDD of 60GB+ but anything less than 60GB has that error, even a Virtual HDD of 59GB won’t work, suggesting FOG could possibly be incapable of deploying a Win10v2004 UEFI image on any drive smaller than 60GB in size unless some code tinkering is done.

  • @george1421 info incoming, it is perfectly possible, this is just a new issue pertaining to v2004, this still works with Non-UEFI Win 10 v 2004, and with UEFI on any v1909 or older Windows 10 release, so this suggests HDD size can’t be the issue. I’ll get some images up shortly tho, thank you for the quick response!

  • Moderator

    @ProfDrSir You are going to need to provide a bit more information than what you have given so far. What error? Can you provide a screen shot of the error?

    Looking at the minimum requirements for Win10 32GB hard drive is the minimum requirements, so trying to deploy to a smaller than 32GB drive will be not supported by windows. ref:

    Single disk resizable only works well when the 😄 drive partition is the last partition on the disk. MS does tend to place the recovery partition as the last partition which is not resizable. This causes FOG to not be able to resize the disk to something smaller than the source disk size from my experience.