EPYC-Milan vs EPYC-Milan-v2

[Seed]

Hello,

Where can I find out the difference between EPYC-Milan and EPYC-Milan-v2. I am experimening setting up a open-webui in docker on a vm. with passthrough and want to have as much native CPU available as possible.

 

[_gabriel]

on a vm. with passthrough and want to have as much native CPU available as possible

Why not simply "host" type ?

 

[Seed]

Why not simply "host" type ?

Id like to know the differences between these two types..

 

[Neobin]

https://pve.proxmox.com/pve-docs/pve-admin-guide.html#_amd_cpu_types

 

[aasami]

Why not simply "host" type ?

I've learned "the hard way" why the "host" type is not always the best solution. QEMU states this also in their documentation that "host pass through" may cause issues with live migration.

 

[MarkusKo]

i could not find much information about that except this
the difference seems to be the cache, milan-v2 is for Milan-X processors and they have additional X3D Cache (like AMD's Gaming CPU's), otherwise they seem to be identical.

for virtualzation i would rather use something more generic like x86-64-v2-AES but if you don't need the portablility between different hosts / cpu's you should use host for maximum performance

 

[ucholak]

DOC: https://pve.proxmox.com/pve-docs/pve-admin-guide.html#_amd_cpu_types

EPYC-Milan : 3rd generation of Zen processors
EPYC-Milan-v2 : add missing flags (+vaes, +vpclmulqdq, +stibp-always-on, +amd-psfd, +no-nested-data-bp, +lfence-always-serializing, +null-sel-clr-base)

i am using EPYC-Milan-v2 on Milan architecture.
IHMO V2 is second version of flags, updated to better reflect arch.

 

[ucholak]

for virtualzation i would rather use something more generic like x86-64-v2-AES but if you don't need the portablility between different hosts / cpu's you should use host for maximum performance

better something like lower common denominator of cpu in cluster. to better use faster instructions and mitigations