Migrating entire cluster from intel CPU to AMD Epyc

[KWKirchner]

We are nearing a tech refresh of our ESXi cluster and are going to be moving off of intel Xeon to AMD Epyc.  Due to port constraints, we will need to run through the cluster and swap one host at a time. This means pulling the existing Xeon host out (remove it from NSX and vCenter) and the putting the Epyc host in its place and joining it to NSX and the existing cluster.

I realize that EVC does nothing for us and everything will need to be powered off and cold migrated to the AMD hosts as we go along.  The only thing we can expect to work across hosts in this temporary mixed state is HA.

Has anyone done this and do they have any lessons learned they can share? Our VM's are primarily Windows 2016 and RHEL7, is there any change that bubbles up from the AMD host that makes this an issue?

[Bivas Mustafa]

You are right. EVC cannot do anything for running workloads. HA should be ok. doing it host by host is good, as it allows you to observe the situation and handle issues gradually. No experience with AMD Epyc though. curious about performance. Let us know ... 

[markwpumphrey]

https://communities.vmware.com/t5/VMware-vSphere-Discussions/Bluescreen-booting-Windows-VM-with-latest-CU-on-AMD-with-VBS/td-p/2918483/page/2

VBS on Windows servers 2016 & up will not boot

As a workaround you can set your "Guest OS Version" to "Microsoft Windows Server 2008 R2 (64-bit)"

you can even enable VBS during a new install while choosing this OS.

"Edit Settings"
in the "VM Options" tab expand "General Options"
"Guest OS Family" should not change "Windows"
"Guest OS Version" should be set to "Microsoft Windows Server 2008 R2 (64-bit)"

[kastlr]

Hi,

 

you should also keep in mind that a few VMs/Services/applications might run into problems after migration is completed.

Simply because the fact that if you/your users installed SW inside their VMs which did use Intel specific CPU features they might report unexpected behavior when running on AMD CPUs.
And as nobody does know (except the SW vendor) if their apps would have used different install routines on Intel and AMD there's no way to identify affected VMs & services prior the migration.   

But I wouldn't expect this to be a bigger problem.

[KWKirchner]

We did get a couple of AMD Epyc servers in our lab and were able to play with them a bit and add one into our existing intel cluster.  No real surprises, powering VMs off one intel host, migrating them to the Epyc host, then powering them back on worked as expected. Did the same in reverse and then removed the AMD host. No issues with NSX adding/removing the host either.