All products 
What 2 cores on a VPS really mean and what fits on them
"2 vCPU" in a plan description does not mean two physical CPU cores. It means two virtual processors that the hypervisor schedules in turn onto the threads of a real server. How much computing power stands behind them depends on the CPU model, the host's policy and your neighbours on the node. That is why two plans with the same "2 cores" line can differ in speed several times over.
Core, thread and vCPU
A modern server processor usually runs two threads per physical core (Hyper-Threading on Intel, SMT on AMD). Hosts almost always count vCPUs by threads. So your 2 vCPUs are at best two threads of one physical core, and more often threads of different cores that other customers need at the same time. Two threads of one core share its execution units and cache, so together they deliver noticeably less than two full cores.
This isn't always visible inside the system: nproc will show 2, and lscpu will print the CPU model and whatever topology the host chose to expose to the guest. Topology in a virtual machine can be nominal, so the real picture comes from benchmarks and watching the load.
Shared and dedicated cores, steal time
- Shared vCPUs are the standard for mass-market plans. Several virtual processors belonging to different customers sit on one physical thread. While the neighbours are idle you get full speed; at peak hours, less.
- Dedicated vCPUs are pinned to your machine, and neighbours have no claim on them. They cost more, but performance is predictable.
- Per-core usage caps. Some plans explicitly limit usage, for example "up to 50% of each core" or "short bursts only". Look for this in the small print of the description and in the host's terms: constant 100% load on such a plan can end in throttling or a warning.
The main indicator of an overloaded node is steal time: the share of time when your vCPU is ready to work but the hypervisor gives the physical processor to someone else. Look at the st column in top or vmstat 1. A few per cent under load is normal; a steady 10% or more means you are effectively getting less than two cores. To compare plans against each other, it is handy to run sysbench cpu --threads=2 run at different times of day and compare the result with a single-threaded run: on honest two cores the difference is close to twofold.
What fits on two cores
Two vCPUs are a comfortable minimum for a server that does more than one thing. Typical scenarios:
- A website or small shop on nginx, PHP-FPM and MySQL with a few thousand visitors a day, provided caching is enabled.
- Several Docker containers: an application, a database, a reverse proxy and monitoring. More on requirements: VPS for Docker.
- Bots and background jobs: a couple of bots, a task queue and a scheduler share two cores comfortably.
- A corporate VPN for a small team or a private network between offices: WireGuard encryption puts only a moderate load on the CPU.
- Builds and tests of small projects, if you're not in a hurry: CI on two cores works, but slowly.
Things get tight when building large projects, running a busy database with many concurrent queries, game servers with dozens of players, and services on Windows. In these cases look at plans with 4 cores. Keep the reverse in mind too: single-threaded programs (a Node.js application, many game servers, some scripts) gain almost nothing from a second core; clock speed matters more to them. There is a selection of VPS with high CPU clock speeds for them.
What to pair two cores with
The CPU rarely limits a server on its own. A sensible pairing for two vCPUs is 2 to 4 GB of memory: with less, the system will start swapping before it loads the CPU; what fits into that amount is covered in detail on the VPS with 2 GB RAM page. For a database, a fast NVMe disk matters more than an extra core: PostgreSQL and MySQL on a slow disk sit idle waiting for I/O, which shows up as a high wa value in the same top while the CPU is almost free. Before buying, check whether the CPU model is stated: two cores of a recent generation are usually faster than four cores of an old server chip, and a single-threaded benchmark in the first days of the rental will show it.