Most server listings quote CPU model and core count up front, but the number of sockets — one CPU or two — changes the RAM ceiling, the price, and even how efficiently your software runs. Picking the wrong one means either paying for capacity you'll never use, or hitting a wall a year into deployment.
A dual-socket (2P) server doesn't just double the core count — it roughly doubles the maximum RAM capacity (each CPU brings its own set of memory channels) and total PCIe lanes for expansion cards, drives and networking. A single-socket (1P) server is simpler, uses less power, costs less to license per physical CPU on software that charges per-socket, and for most everyday workloads is genuinely enough.
Dual-socket servers use a NUMA (Non-Uniform Memory Access) design: each CPU has its own local memory bank, and reaching the other CPU's memory over the inter-socket link is slower than reaching local memory. Software that isn't NUMA-aware — some older applications, and workloads with a lot of cross-thread memory sharing — can actually run slower on a busy dual-socket box than on a single-socket one with the same total core count. It's not a reason to avoid dual-socket, just a reason not to assume "two CPUs = automatically faster" for every workload.
A refurbished dual-socket server is only worth the premium if both CPU sockets are actually populated and matched — a 2P chassis running with one socket empty gives you the power draw and cost of a bigger box without the benefit. Confirm both CPUs are installed, matched in model/stepping, and that the memory is populated evenly across both CPUs' channels (uneven population quietly disables the performance benefit of the second socket). Every server we sell lists exact socket count and CPU population up front, tested and backed by a 3-year warranty — browse current refurbished servers or check indicative price ranges by configuration.