UK’s trusted IT infrastructure partner since 2003
Servnet
ConfiguratorGet in Touch
Intel Xeon 6 for UK server buyers: P-cores, E-cores and refresh timing — analysisIntel Xeon 6 for UK server buyers: P-cores, E-cores and refresh timing — analysis — reach
Server Infrastructure · Trends

Intel Xeon 6 for UK server buyers: P-cores, E-cores and refresh timing

Servnet Editorial · Server Infrastructure Practice12 min read

Intel Xeon 6 is the first server generation in years where the headline question is not how many cores you can buy, but which kind of core you want. The line splits into Performance-cores for latency and per-core throughput and Efficiency-cores for raw density at low power, and the two families now ship as distinct products rather than a single ladder. For UK buyers weighing a refresh, that split changes how you size a host, how you budget for per-core software, and whether this is the generation to skip a cycle. Here is what the P-core and E-core divide actually means when you are signing the purchase order.

Xeon 6 P-core vs E-core vs prior gen
P-core 6900P/6700PE-core 6900E/6700EPrior-gen XeonCore typePerformanceEfficiencySingle ladderCores/socketModerate, fastVery highLowerPer-core perfHighestLowerBaselinePerf-per-wattStrongHighestLowerBest useDB / VMsScale-outRefresh from

Two kinds of core, two kinds of server

Earlier Xeon generations gave you one architecture and asked you to choose a point on a single core-count-versus-clock curve. Xeon 6 separates that curve into two product families on the same platform. The P-core parts (the 6900P and 6700P series) carry Intel's full-fat Performance cores with the widest per-core execution and the strongest single-thread clocks, which is what databases, virtualisation hosts and most general business workloads actually feel. The E-core parts (the 6900E and 6700E series) pack a much larger number of smaller Efficiency cores into the same socket, trading per-core punch for thread count and performance-per-watt.

The practical effect is that you now pick the family before you pick the SKU. A consolidation or database host wants P-cores: fewer, faster cores that finish work quickly and keep per-core licensed software in check. A scale-out tier built from many small stateless instances - web front ends, microservices, CDN nodes, certain analytics fan-outs - is where E-cores earn their keep, because the workload scales with thread count and is not throttled by any single thread's speed.

Why the split matters for per-core licensing

The most expensive thing on many servers is not the silicon - it is the software licensed by core that runs on top of it. VMware under Broadcom is per-core subscription with a 16-core-per-CPU minimum; SQL Server Enterprise, Windows Server Datacenter and Oracle are all core-based. On a P-core host, the discipline is the same as it always was: buy the fewest, fastest cores that meet the workload, because every core is a recurring licence line as well as a one-off purchase.

E-cores invert that logic and only make sense where the software stack is not licensed per core - open-source web and container workloads, internal services, build farms. Dropping a 144-core E-core part into a per-core-licensed estate would be an expensive mistake. Decide the licensing model of what runs inside the box before you let core count climb, exactly as we set out in how to spec a server in 2026.

  • P-cores (6900P / 6700P): databases, virtualisation, general business apps, anything per-core-licensed
  • E-cores (6900E / 6700E): stateless scale-out, web/container fleets, build and CI, performance-per-watt density
  • Mixed estate: P-core hosts for the licensed tier, E-core hosts for the open-source tier

Platform gains beyond the cores

Xeon 6 is not only a core story. The platform moves to more memory channels of DDR5 and supports the new MRDIMM standard, which lifts usable memory bandwidth - the metric that actually limits analytics and large virtualisation hosts more often than core count. It carries more PCIe 5.0 lanes plus CXL, so you can attach more NVMe, more accelerators and, in time, pooled or expanded memory. For bandwidth-bound workloads the uplift over a prior-generation Xeon can be larger than the core numbers alone suggest.

That extra I/O and memory capacity is also why a one-for-one refresh often consolidates more than expected: a single Xeon 6 host can absorb the role of two older nodes once you account for memory bandwidth and lane count, not just cores. Size memory deliberately - balance every channel - using our memory and RAM guidance, because an unbalanced layout strands the bandwidth you paid for.

Consolidation upside over a refresh
967248240Y0Y1Y2Y3Y4Years after refreshCumulative benefit (index)Perf-per-watt savingLicence + rack saving

Refresh timing: skip a cycle or jump now?

If your fleet is on Cascade Lake or Ice Lake era Xeon, Xeon 6 is a genuine generational jump and the consolidation maths usually pays for itself in power, rack space and licence reduction. Fewer, denser hosts mean fewer sockets to licence and fewer kilowatts to cool - which, given UK energy prices, is a real line item rather than a rounding error.

If you are on the immediately preceding Sapphire Rapids or Emerald Rapids generation, the calculus is closer. There the question is whether the bandwidth, CXL and density gains justify a mid-life refresh, and that depends on whether you are bandwidth-bound, power-constrained or chasing consolidation to cut software cost. The honest answer is workload-specific, which is why we model it per estate rather than by rule of thumb.

Putting it together for a UK buyer

Start from the workload and its licensing, choose the P-core or E-core family accordingly, then size memory channels and PCIe before you fixate on the core number on the box. For most UK businesses the answer is a P-core 6700P or 6900P part in a mainstream dual-socket PowerEdge, ProLiant or ThinkSystem, with E-core parts reserved for genuine scale-out tiers. Pick the exact processor with our processors guidance and build and validate the full host in our server configuration service.

Key takeaways
  • Xeon 6 splits into P-core (6900P/6700P) and E-core (6900E/6700E) families - choose the family before the SKU.
  • P-cores suit databases, virtualisation and anything licensed per core; E-cores suit open-source scale-out density.
  • Per-core software cost, not silicon, should cap your core count on licensed hosts.
  • Bigger memory bandwidth (more DDR5 channels, MRDIMM) and more PCIe 5.0/CXL lanes often matter more than core count.
  • Refreshing from Ice Lake era pays off on consolidation, power and licences; from Sapphire/Emerald Rapids it is workload-specific.
Frequently asked

FAQs — Intel Xeon 6 for UK server buyers

P-cores vs E-cores

Should I buy Xeon 6 P-cores or E-cores?

P-cores for databases, virtualisation and anything licensed per core - they give the strongest per-core clock. E-cores only where the stack is not per-core-licensed and scales with thread count, such as web and container fleets. We pick the family per workload in our configuration service.

Do E-cores hurt per-core software licensing?

They can. A high-count E-core part on per-core software like VMware or SQL Server multiplies licence cost for cores you may not saturate. Reserve E-cores for open-source scale-out tiers and keep licensed workloads on fewer, faster P-cores per our processors guidance.

Refresh timing

Is Xeon 6 worth a refresh from older Xeon?

From Cascade Lake or Ice Lake era, usually yes - consolidation cuts sockets to licence and kilowatts to cool. From Sapphire or Emerald Rapids it is closer and depends on whether you are bandwidth- or power-bound. We model it against your estate in server configuration.

Related

Got a question this article didn't answer?

One conversation with an engineer who's done this before. No sales script.

Talk to Servnet →