For two decades the default enterprise server was a dual-socket box, and most buyers still tick that option without thinking. That habit deserves a fresh look. Modern single-socket Intel Xeon 6 and AMD EPYC processors now offer core counts, memory channels and PCIe lanes that used to require two CPUs, which means a single-socket server is no longer a compromise for many workloads, and is sometimes the smarter buy on cost, power and licensing. This is the framework we use with UK customers to decide whether the second socket earns its place, or whether you are paying for capability you will never use.
Why dual-socket became the default
The second socket exists to add resources: more cores, more memory channels and more PCIe lanes than a single CPU of its era could provide. For years that mattered, because a single processor simply could not host enough cores or address enough memory to run a busy virtualisation cluster or a large database, so two sockets were the only way to reach the required scale in one chassis.
That logic still holds at the top of the range, but the floor has moved. A modern single processor now carries enough cores, enough memory channels and enough lanes that, for a great many workloads, the resources of one CPU are sufficient on their own. The question is no longer whether two sockets are more capable, because they are, but whether your workload actually needs that extra capability or is simply inheriting an old assumption.
What the second socket actually costs
A second socket is not free capability. It adds the cost of the CPU itself, more DIMM slots to populate to keep its channels balanced, a higher idle and loaded power draw, and a larger cooling and thermal envelope. On a server that never uses the extra cores, that is money and watts spent on resources sitting idle, and over a multi-year life the power difference alone is a real line on the bill at current UK energy prices.
There is a licensing angle too. Software licensed per core, including some hypervisor and database products, is indifferent to how many sockets the cores sit in, but a needlessly large two-socket box invites you to fill it with cores you then pay to licence every year. Right-sizing to a single capable socket can lower both the hardware bill and the recurring licence spend, which is the same core-count discipline we describe in how to spec a server in 2026.
- •Second CPU plus more DIMMs to keep its channels balanced
- •Higher idle and loaded power, and a larger thermal envelope
- •Per-core software cost rises if you fill the extra sockets needlessly
- •Unused cores are capital and watts spent on idle capacity
When a single socket is the right answer
A single-socket server is the better buy for a long list of common roles: edge and branch servers, dedicated application or web hosts, domain controllers and other lightweight infrastructure, smaller virtualisation hosts, and dense hosting where you want more, smaller servers rather than fewer large ones. In all of these the resources of one modern CPU are ample, and avoiding the second socket keeps the platform cheaper, cooler and simpler.
Single-socket also sidesteps a subtle performance trap. With one processor there is no remote-memory penalty, because there is no second CPU whose memory a thread might have to reach across the interconnect to use. For latency-sensitive single-application servers that simplicity is a genuine advantage, removing a class of tuning problems that dual-socket boxes create. Pick the exact processor for the role with our processors guidance.
When you genuinely need two sockets
The second socket still earns its place where the workload truly needs the aggregate resources of two CPUs in one box. Dense virtualisation consolidation, where you are packing as many VMs as possible onto each host, benefits from the combined cores and the larger memory footprint two sockets allow. Large in-memory databases that must address enormous amounts of RAM need the memory channels and DIMM slots only a second socket provides.
PCIe-hungry builds are the other clear case. A server loaded with multiple GPUs, several high-speed NICs and many NVMe drives can run short of lanes on a single CPU, and the second socket brings another full set of lanes to feed them. If your build is genuinely resource-bound in cores, memory or lanes, two sockets are correct; the discipline is making sure that need is real rather than assumed.
The NUMA factor
A dual-socket server is a NUMA machine: each CPU owns its own memory, and a thread reaching memory attached to the other socket pays a latency penalty crossing the interconnect. Well-tuned virtualisation and database platforms account for this by sizing VMs and workloads to stay within a single NUMA node, but it is a real consideration that a single-socket server simply does not have, because there is only one memory domain.
This is not a reason to avoid dual-socket where it is needed, only a reason not to choose it casually. If you do buy two sockets, populate both symmetrically so each NUMA node has equal capacity and bandwidth, and size workloads to stay local where latency matters. If your workload is a single latency-sensitive application, the absence of NUMA on a capable single socket is one more point in its favour.
Putting it together
Start from the workload, not the habit. If the role is a single application, an edge or branch box, a lightweight infrastructure server or a smaller virtualisation host, default to a single capable socket and only step up if you can name the resource you would run short of. If the role is dense consolidation, a large in-memory database or a lane-hungry GPU build, the second socket is justified. Either way, build and compare both as real configurations in our Dell configurator and let the spec, power and cost decide.