Just crowdsourcing some info, but for those of you with non-persistent, multi-session VDAs hosted in Azure, what size machines are you running? We've been testing with Standard_D16as_v5, but are also considering something in the F-series.

D8as or d8ads is a good starting point, so similar to the D16as.

We use Standard_D8as_v4 as well as Standard_D4as_v4 depending on the resource and what they are used for. Both are running Windows 11 Multi-Session.

Standard_D16ds_v5 and Ephemeral disk.

Started with D8asv4, but switched to NC8as_T4 due to GIS apps (and Office/Teams) behaving so much nicer with the GPU backing.

In our LogonVSI testing, we were definitely CPU bound. This is all S16 BTW (I know, I know...).

It seems that LoginVSI tests are cpu bound most of the time unless you really try to get your own apps included but i would say in production environments i normally see multisession vdas being memory bound. But why dont you just look at your existing systems they would be a much better indicator compared wiht a generic test.

This is for Azure NP, so we have nothing being used at the moment.

Very interested in this topic also. I was reading here also that MCSIO with correct RAM would be more performant and cost efficient than even using ephemeral disk. What do you think?

As long as the cache can keep up it would eb because you are using RAM but if it cant then ephemeral would be faster. You could test and use perfmon to monitor

Thanks Rob

@Aaron O'Reilly, do you still benefit from GPUs for Teams if it's HDX optimized?