[SOLVED] Can I move VMs from Intel to AMD?

[Razva]

Currently I have a bunch of VMs that run on a E5-2630 v3 (12k CPU Benchmark points) for which I'm paying way too much.

I found a i7-8700 (15k CPU Benchmark points) at a good price, and a Ryzen 5 3600 (19k CPU Benchmark points) at an even better price.

The main question would be: is it "safe" to move VMs from the Intel platform (E5-2630) to a AMD platform (R5 3600), or will VMs break because of random "incompatibilities"?

Thank you!

 

[aaron]

Live migration between two nodes with very different CPUs can fail.
Shutting the VM down and then booting it up on the other CPU should work.

Depending on how you configured the CPU for the guest (if "host") a Windows install might complain about being activated again because the hardware changed considerably.

 

[Razva]

There will not be any live migrations. VMs will be fully stopped, migrated and started again.

 

[sseidel]

We do Live Migrations between AMD and Intel without problems. The key is for Windows to set the CPU type to "Westmere". Then there's no problem. Cold migration (shut down, move, start) is no problem regardless the CPU setting.

I have never had any version of Windows complain about a different CPU regarding activation. So far I have used VMs with Windows XP, 2000, 7, Server 2008/R2/2012/R2/2016/2019 on at least 10 different CPUs from Intel Xeon 5500 series to AMD EPYC and none complained ever about the CPU change, even with "host" CPU setting. It can be different if you change other settings.

Linux VMs simply don't care at all.

 

[spirit]

it's really russian roulette. depend of amd processors.

not sure with epyc, with I have problems between intel and old amd opteron 63XX. (migration is succefull, but vm crash after some hours).


So you need to test it to be sure it's working fine for you.

 

[Drthrax74]

Just disable drivers with autoruns et modify config

 

[starkruzr]

Has anyone tried this lately? I just added a Zen machine to my cluster and when I migrate VMs to it their consoles stop working.

 

[foxpalace]

Hi,
I have this problem at the moment too - after migratin from AMD to INTEL (VM runs with CPU-Type kvm64) the machine freeze in random time.
What means "autoruns et modify config"?

 

[legbreaker]

Currently I have a bunch of VMs that run on a E5-2630 v3 (12k CPU Benchmark points) for which I'm paying way too much.

I found a i7-8700 (15k CPU Benchmark points) at a good price, and a Ryzen 5 3600 (19k CPU Benchmark points) at an even better price.

The main question would be: is it "safe" to move VMs from the Intel platform (E5-2630) to a AMD platform (R5 3600), or will VMs break because of random "incompatibilities"?

Thank you!

Don’t migrate them live..
Just shutdown the vms, migrate them and boot them up on the AMD kvm host..trust me..

 

[Jannoke]

Live migrating between Xeon(R) CPU D-1541 and AMD EPYC 7401 will randomly crash Centos 8 guest OS. Funny enough, it will not crash straight away. It just freezes guest few minutes after migration has ended. Reset of guest (or stop-start) will reboot and fix it.
It won't crash openwrt 21.02.

 

[Boogers]

I can confirm I have a similar issue. I am Migrating from an Epyc 7742 to and Intel i3-9100F. My Linux ubuntu 20.04 machines crash, but It appears that my PF Sense(FreeBSD) and Windows Server 2019 VMs don't. They are all kvm64.

 

[t.lamprecht]

They key point to take away is still:

it's really russian roulette. depend of amd processors.

not sure with epyc, with I have problems between intel and old amd opteron 63XX. (migration is succefull, but vm crash after some hours).


So you need to test it to be sure it's working fine for you.

amd64/x86_64 is basically just a base of agreed instruction set, different CPU implementations can and do add their own extensions and flags, while that part is almost fully controllable through using a common denominator like kvm64, but the architecture differences don't stop there at all, especially for models of different vendors. Slight differences in how the memory model works, how side effects show up, and so on, can cause critical sections to fail as parts of the program was set up under different conditions that it continues to run.

For production loads always try to make a cluster as homogenous as possible, different CPU models from the same series are normally just fine, different series from the same vendor is most often fine too, at least if they're not ages apart and as long as the VM is configured to use the greatest common divisor model between those. Different vendors: there be dragons.