does proxmox support virtualization of hyper threading (using multi threads available on the same CPU)?
Hyper threading virtualization
- Thread starter Maher Khalil
- Start date
Not sure what you actually mean.
I think there is no GUI option to create VMs with multiple threads per vCPU like other KVM hypervisors do.
But several vCPUs can use the same physical CPU core, because Proxmox will hypervise and spread multiple threads across all the physical CPU cores/threads. And Proxmox can make use of the hyperthreading of your physical CPU.
I think there is no GUI option to create VMs with multiple threads per vCPU like other KVM hypervisors do.
But several vCPUs can use the same physical CPU core, because Proxmox will hypervise and spread multiple threads across all the physical CPU cores/threads. And Proxmox can make use of the hyperthreading of your physical CPU.
I am trying to get releation between threads and vCPU like this
vCPU numbers = No. of CPUs x No. of cores per CPU x No. of threads
vCPU numbers = No. of CPUs x No. of cores per CPU x No. of threads
Using Proxmox you can allocate more vCPUs than you actually got physical CPU cores if you mean that. So if you for example got a 8 core 16 thread CPU you can easily allocate 64 vCPUs as long as the VMs using these vCPUs are idleing most of the time.
The point is I would like to calculate the host resource to purchase to provide 62 nos. of 4vCPUs VMs. load may be 50%
so, how can I decide no. of sockets, cpus, and threads?
so, how can I decide no. of sockets, cpus, and threads?
62*4 @50%= 124.
to do it on a single system you need a 4 socket machine- easier/more cost effective to split it between multiple hosts. OR you might want to rethink your load assumption.
what about intell 1 socket x 12 core per socket x 24 thread per core = 288 vCPUS (If I am right)
will it have good performance?
or I have to do another calculations?
will it have good performance?
or I have to do another calculations?
Last edited:
there's no cpu that takes 24 threads per core.
Both intel and AMD hyperthreading provides 2x per core BUT you dont get the performance of 2 cpus for that. depending on what you feed it you can get 100-140% of a single core performance across both. if you're looking for guaranteed performance, ignore vCores (the logical cores resulting from hyperthreading) and count just actual cores.
Both intel and AMD hyperthreading provides 2x per core BUT you dont get the performance of 2 cpus for that. depending on what you feed it you can get 100-140% of a single core performance across both. if you're looking for guaranteed performance, ignore vCores (the logical cores resulting from hyperthreading) and count just actual cores.
for example the below mentioned new intel processor
Intel® Xeon® Silver 4310 2.1G, 12C/24T, 10.4GT/s, 18M Cache, Turbo, HT (120W) DDR4-2666
the part 12C/24T
as per my understanding is
12C = 12 cores
24T = 24 thread per core
am I right?
check this link
https://www.intel.com/content/www/u...sor-13-75m-cache-2-20-ghz/specifications.html
Intel® Xeon® Silver 4310 2.1G, 12C/24T, 10.4GT/s, 18M Cache, Turbo, HT (120W) DDR4-2666
the part 12C/24T
as per my understanding is
12C = 12 cores
24T = 24 thread per core
am I right?
check this link
https://www.intel.com/content/www/u...sor-13-75m-cache-2-20-ghz/specifications.html
Last edited:
understand it now
Thank you very much
so, how can plan the host based on the my vCPU requirements
if I need 124 vCPUs as stated above, how can I plan the host?
if 1 socket so
124 vCPU = 1 socket x no. of cores x 2 threads
so,
no. of cores = 124/2 = 62 cores
or no. of cores = 32 in 2 sockets
or no. of cores = 16 in 4 sockets
right?
Thank you very much
so, how can plan the host based on the my vCPU requirements
if I need 124 vCPUs as stated above, how can I plan the host?
if 1 socket so
124 vCPU = 1 socket x no. of cores x 2 threads
so,
no. of cores = 124/2 = 62 cores
or no. of cores = 32 in 2 sockets
or no. of cores = 16 in 4 sockets
right?
Last edited:
you need 124 cores. you might be able to get away with ~180 vcores, but I humbly suggest your assumptions are wrong.
Why/how did you decide that's how much CPU you actually need? its completely possible you intend to deploy 62 vms with 4 cores each, but that doesn't mean they're all gonna consume 50% cpu at the same time. You need to model your use case and arrive at a maximal load at peak time; then you can decide to build out your infrastructure to provide that level of load, or depending on budget and resilience/latency sensitivity to a lower mean.
In any case, scaling to 62 vms is pretty simple- add nodes until you're covered. remember to leave sufficient compute/ram overhead to your hypervisor os (eg proxmox) and any storage infrastructure.