Specific question about machines and hyperthreaded pairs.

[DarkZenith]

I have the parts to throw together a machine for my kids and for my nas/plex. 

 

It is a dual Xeon e5-2670 on a Supermicro Mainboard with ipmi (Remote network desktop/bios for unraid control and NAS access. So the machine will have 16 cores and 32 threads.  It will also have 128gb of ram.  I also have x2 nvidia gtx 1060s and a gtx 980ti for the passthrus for my kids machines.

 

Here are my specific questions I want to get ironed out so I can finish my plans.  

 

1) Cores when you have a typical i7 quad core are paired as (0,4) (1,5) (2,6) (3,7) for the physical and hyper threaded cores. In a dual cpu configuration does it do physical processor cores 0-15 and then HT cores as 16-31. Or are they split by processor? eg cpu 0 is (0-7, 8-15) and cpu 1 is (16-23, 24-31)?

 

2) the second question is about how I am planning to setup the kids  processor assignments.  I plan to give each x4 physical cores and x4 then cores for their machines.  I was wondering if I did a funky core assignment like the following (assuming cores 0-15 are the physical ones)

system 1, cores 0-3, 20-21, 24-25

system 2, cores 4-7, 16-17, 26-27

system 3, cores 8-11, 18-19, 22-23

system 4, cores 12-15, 28-31

System 4 is for unraid/nas functions, plex and such.  

The reasoning behind my core assignments is like this, each kid has x4 physical cores assigned to them, and x2 virtual cores assigned from the other 2 machines.  The idea is that when the other machines are in light load or not in use that would give the matching virtual core near 100% performance available for those unused cores.  End result is instead of 4 cores with 4 matching ht cores only effectively allowing 50% of the cpu resource for those cores each could potentially allow a full or near full 100% per core (as I said IF the matching physical core isn’t under load).  

 

Would this idea/concept work or is it just a paper theory that doesn’t pan out?

 

3) Would there be any issue of assigning each machine 2 cores off each physical processor (to split the load) of would it perform better keeping them on one die?

 

I’ve been sitting on these parts for nearly a year now and it is time to finally put them together.  Any advice would be appreciated.  

 

PS - a quick question, has USB port passthru gotten easier to configure yet as I have x3 cloud 2 headsets that would be in use. Also, can the ps/2 ports be hardware passed through or are those different because of the way they use direct bus access? (I have an x-arcade I was hoping to use as there is zero latency on ps/2 with it.

[SSD]

Generally it is advised to assign full cores to a VM, as the splitting of work between a core and its pair is best handled writhin one VM.

 

Note that the same cores can be assigned to multiple VMs so that they are sharing. 

 

Have not seen anyone running tests to try to optimize a shared CPU model, or a split core model, but would be interesting to see that.