When ordering a VPS, you can use select a VPS with, amongst other specifications, shared or dedicated CPU cores. The difference between these two is that with shared CPU cores you use the same CPU cores of a hypervisor as other VPSs. With dedicated CPU cores, the cores are reserved for you, regardless of whether you fully utilize them or not.
We illustrate this with an example scenario: We assume a hypervisor with 50 CPU cores and 100 CPU hyperthreads. We then only place VPSs with 4 CPU threads each on the hypervisor.
-
Dedicated CPU cores: The number of VPSs on a hypervisor is the number of CPU cores of that hypervisor, divided by the number of available CPU cores per VPS. In this example, there could be 100 (threads) / 4 = 25 VPSs on this hypervisor.
-
Shared CPU cores: You share the cores of a hypervisor with other VPSs. Suppose that each VPS on a hypervisor uses an average of 25% of its available CPU cores, then there can be four times as many VPS's on one hypervisor as the actual number of cores, so that would be 100 (threads) / 4 (cores) * (100/25) = 100 VPSs.
Additional reading tip: Shared cores can play a role in the emergence of steal-time and is one of the reasons why we use automatic migrations on our VPS platform to balance the CPU load across hypervisors.
To summarise, more VPSs with shared CPU cores can be on one hypervisor at the same time than VPSs with dedicated cores. This also plays a role in the price of Windows Server licenses, where license prices are based on the amount of cores of the actual hypervisor. A shared CPU platform allows spreading these costs over a larger amount of VPSs than a dedicated CPU platform.
When do you use dedicated CPU cores instead of shared CPU cores?
Because dedicated CPU cores are reserved entirely for you, the price is slightly higher than with shared CPU cores. You might therefore wonder when you actually benefit from dedicated CPU cores compared to shared CPU cores.
Simply put, we recommend using dedicated CPU cores for all AI applications and remote desktop servers. For example, using a custom AI model can generate tokens approximately 3-4 times faster with shared CPU cores. B
For database servers, dedicated CPU cores can make a difference, but this depends on the load on your database. In this case (and all other scenarios), we recommend using the regular VPS because it can flexibly switch between dedicated and shared CPU cores. If you find that you need more performance, you can simply test dedicated CPU cores and decide whether or not to deploy them permanently based on your findings.