I see this on Proxmox with 2025 and 11. I have some instances at NetCup (Hoster using KVM), same issues but with the disadvantge that i cant play with host/cpu settings. All on AMD Epyc 96xx
AMD : BSOD unsupported processor since Windows build 26100.4202+ ( update kb5060842 + its preview kb5058499 )
- Thread starter KciNicK
- Start date
I was bitten by the same issue yesterday evening. KB506842 fails to install, with an Unsupported processor error after at the reboot stage.
Seems KB506842 has been superseded by KB5063060 today.
I can report KB5063060 does not resolve the issue. The aforementioned workaround, setting the CPU type to
Proxmox 8.4.1, host is a 5950X and Windows 11 24H2 guest.
Seems KB506842 has been superseded by KB5063060 today.
I can report KB5063060 does not resolve the issue. The aforementioned workaround, setting the CPU type to
EPYC does allow the update to install, and the system to run.Proxmox 8.4.1, host is a 5950X and Windows 11 24H2 guest.
One of my computer using a 5800X3D suffers from the same issue since the preview patch, but my 5950X doesn't for reasons...
Moreover on my 5800X3D one of my VM suffers from the boot loop but the other one, which is basically the same one but with different mem/proc count and a different video card passed through works fine even with patch KB5063060 on CPU host while i had to change cpu type to Epyc for the other one...
Moreover on my 5800X3D one of my VM suffers from the boot loop but the other one, which is basically the same one but with different mem/proc count and a different video card passed through works fine even with patch KB5063060 on CPU host while i had to change cpu type to Epyc for the other one...
Attachments
is VM on Windows 26100 build ?
is VM cpu type "host" vanilla or custom args in VM .conf ?
Host is vanilla, no change in the conf file.
This one is 26200.5570 not 26100 it used to be my physical dev machine which i VMized, it works fine. Forgot it was on insider channel.
There is still the issue of my second 5800x3d VM which is working fine on 26100.4351 while another VM on the same host doesn't.
Attachments
26200 is based on 26100
but 26200.5570 is April Dev Channel update.
June Dev Channel build is 26200.5641 which has same problem as 26100 Normal Channel since 26100.4202
I don't recommend this, as performance degrades significantly (in particularly for IOMMU GPU passthrough). Nevertheless it is worth mentionining to get to a better solution.
Enabling the Virtual Machine Platform feature within the VM works, as suggested on this Reddit post:
https://old.reddit.com/r/VFIO/comments/1l98f2k/amd_processor_unsupported_processor_bsod_after/
1. Install KB5063060 with the CPU set to
2. Enabling the Virtual Machine Platform feature within the VM:
3. Reboot, to fully install the Virtual Machine Platform feature, and then shutdown.
4. Change the CPU back to
From a performance perspective, there isn't much architectural change between Zen, Zen 2 and Zen 3--they all support AVX2/FMA3. (There are many microarchtectural change between the Zen generation, but that is largely a hidden implementation detail.) (Zen 4 brings in major change with AVX512.)
Leaving the CPU set to
In my case my host is a 5950X (Zen 3). For example, a 5950X host will lose the FAST SHORT REP MOVSB feature. (Which is of questionable value, see https://lunnova.dev/articles/ryzen-slow-short-rep-mov/)
Enabling the Virtual Machine Platform feature within the VM works, as suggested on this Reddit post:
https://old.reddit.com/r/VFIO/comments/1l98f2k/amd_processor_unsupported_processor_bsod_after/
1. Install KB5063060 with the CPU set to
EPYC-Rome-v4 (Zen 2)2. Enabling the Virtual Machine Platform feature within the VM:
3. Reboot, to fully install the Virtual Machine Platform feature, and then shutdown.
4. Change the CPU back to
hostFrom a performance perspective, there isn't much architectural change between Zen, Zen 2 and Zen 3--they all support AVX2/FMA3. (There are many microarchtectural change between the Zen generation, but that is largely a hidden implementation detail.) (Zen 4 brings in major change with AVX512.)
Leaving the CPU set to
EPYC-Rome-v4 (Zen 2) doesn't have the same performance degradation. However that does effectively mask off CPUID Zen 3 specific feature bits.In my case my host is a 5950X (Zen 3). For example, a 5950X host will lose the FAST SHORT REP MOVSB feature. (Which is of questionable value, see https://lunnova.dev/articles/ryzen-slow-short-rep-mov/)
DOHHHH ! That's the reason why my second 5800X3D VM works fine ! I use it for programming purposes and it has virtualization enabled for VS Code !
When I looked into this for my 5950X and 5700G hosts, Rome-v3 is the recommendation as v4 is for Zen4 and these CPUs are Zen3, this was based on:
it's important to note that the Ryzen 9 5950X supports the CPU type EPYC-Rome-v3 but not EPYC-Rome-v4. This is because EPYC-Rome-v4 includes features that are not supported by the 5950X, such as xsaves, which were removed in a recent AMD microcode updateSo is it better to use v4 then?
I really hope this gets fixed soon.
it's better to use v4, if it's not supported (the feature is not there), the vm will not start, then use the v3..., if v3 don't start, v2, so on...
I had thought Rome was Zen 2, and Milan was Zen 3.
It's not clear to me the precise differences between Rome-v1 through v4. I was looking for clarification on that, but it doesn't seem to correspond to Zen 1-4.
I believe the 5950X itself does support XSAVES, as the 1386 errata applies to Zen 2.
In either case, my 5950X host boots as CPU model EPYC-Rome-v4. It's my understanding that libvirt sanity checks the CPU feature level, and will fail to boot a guest if the host CPU CPUID bits aren't a superset of the guest CPU model.
Last edited:
Ok, I'll try v4 on my test machine and see if it still boots - I'd rather be running the better one while we wait for a fix.