>There is 12 Vcores for this VM when all 20 users are pounding the system average CPU is 4% to 17% They can be some peek to 100% now and then on some Vcores.
>Total amount of data is about 20 Gig and 6 gig are contains in transactions tables that are access all the time: Purchase, Reception, Sales Order, Invoice...
>
>I don't have a specific issue but when I installed on that server I got a small gain in performance but I didn't fell off my chair.
>I was wondering if it was Foxpro that needed some tweaking with some settings...

OK, if the VM has 48GB RAM and you allocate 1GB each to 20 users (=20GB), that leaves 28GB for other things.

The VM will use a lot of that for disk cache by default. If your ERP app is read-mostly, with not a lot of writes, that will benefit a lot from the disk cache, even if write caching is disabled.

With a total of 20GB of tables, of which 6GB are actively used, I'm guessing the entire data set is being cached in RAM. Once the first user loads them up in her ERP session, they're already cached for all subsequent users in other sessions.

The initial data load will be faster from the PCI SSD but once in the cache the server wouldn't actually care too much if the persistent storage was the SSD or spinning rust. So the sustained performance of the app over time might depend more on individual core speed and total memory bandwidth. You might not have got too much of a bump in performance in those parameters, when moving to your current server.

It's worth knowing how many physical cores you have on Intel processors, especially Xeon server processors. A lot of them implement HyperThreading, which can get more performance out of a single core through advanced scheduling tricks and presenting a single physical core to the OS as 2 "logical" cores.

However, this doesn't double the performance - as a rule of thumb you get about a 30% improvement. You might see a Xeon processor that has 8 physical cores w/HT (often marketed as 8C/16T). Real-world performance of that chip would be that of about 10.4 physical cores.

Hyper-V vcores only look at logical processors. So, if the server supports HT and you've allocated 12 vcores, that's only equivalent to about 7.8 physical cores (6 * 1.3). Obviously that's still plenty if you're only seeing a peak of 17% global CPU utilization under load.

If ERP generates significant write traffic to temp or other files, it might be worth experimenting with a RAM disk. That way you could leave overall disk write caching disabled but still benefit from some or most of your writes never leaving RAM. Any RAM you allocate for one would then become unavailable for disk cache or other use by the VM.