GHz or core count?

cvega

Member
Oct 30, 2019
13
3
8
45
I know this question can be asked in a million ways, and probably has a million variants of an answer, so maybe someone can simplify this for me ;)

Host will run approx 10 vms, mix of windows / linux. Storage is a mix of spinning and NVME drives attached to a dell Perc H310 controller in various guises of RAID.



faster clock with less cores (i7-9700) or slower clock with more cores (e5-2650 v4) ?
 
If not all VMs are busy at the same time, an 8-core/16-thread CPU will probably be good enough, which is available as a consumer platform (i7 or Ryzen 7, which are cheaper than Xeon or Epyc). I suggest you get the fastest one that your budget allows, including the motherboard (which I guess you also have to buy because of the different sockets for your two options).
Maybe get an AMD 2700X really cheap now (with ECC memory), and later upgrade to an AMD 5900X or better on the same X570 motherboard?
PS: If you need many PCIe lanes for PCI passthrough or a lot of memory, you might want to go Xeon or Threadripper.
 
Last edited:
My server runs 4 kvms + 8 lxc containers on an old dual socket xeon (total 12c/24t @2.8G) and my peak CPU usage during the last month was 25% and is more typically 10-15%

My gut feeling is that VM performance is generally more dependent on RAM and IOPS than CPU grunt but then I don't tend to run anything particularly CPU intensive so your mileage may vary.

Containers seem to run at near native speeds, certainly no complaints there at all.

Windows VM's are a bit slower but I find that giving them enough ram and making sure they're hosted on the fastest storage with the right virtual drivers is the best recipe for decent performance.
 
I did look at AMD, the issue i have that they have no igpu, which wastes a slot which i want to use for Dell H310 / 10gbe card. I am constrained to a microatx board at this time.

At the moment, I'm running a 2650v3 with 64gb ram, and i find performance mostly good (10c/20t), but i have to move to a microatx platform due to space constraints.