P Core vs E Core and Intel Hybrid CPU Architecture Explained

P-core vs E-core describes the two kinds of CPU core in Intel’s hybrid processors: performance cores (P-cores), built for fast single-thread work, and efficient cores (E-cores), built to deliver the most work per watt. Intel introduced the design with its 12th Gen Alder Lake processors, kept it through the 13th and 14th Gen Raptor Lake parts and every Core Ultra series since, and added a third type, the low-power E-core (LP E-core), from Core Ultra Series 1 onwards.

The idea grew out of a long trend in embedded computing. Since processors began to take shape in the 1980s, each step has met new industrial demands: by around 2000, embedded applications needed to multitask, which pushed Intel towards dual-core and then multi-core processors, and the Core series brought quad-core designs and Hyper-Threading for real-time monitoring and control. Adding identical cores raises performance, but it also raises cost, power and heat. A hybrid processor takes a different route: a few large cores for the work that needs speed, and many small cores for everything else.

Hybrid processors first reached our Intel industrial computers with the 12th Gen processors, codenamed Alder Lake. This guide defines each of the three core types, compares them side by side, sets out how the mix has changed by generation, explains how to read core counts and how the operating system schedules work, and covers what hybrid cores mean for real-time, virtualised and fanless industrial systems.

The table compares the three core types as they appear across Intel’s hybrid processors. Exact clocks and cache sizes vary by generation and model.

All three run the same x86 instructions, so software does not need to know which type of core it is on; the difference is speed and power. The one exception in practice was AVX-512, which Intel disabled on hybrid Alder Lake parts so that every core supported the same instruction set.

A P-core, or performance core, is the large, high-clock core in an Intel hybrid processor, built to finish single-threaded and latency-sensitive work as quickly as possible. P-cores work like the cores in Intel’s earlier multi-core chips. They are tuned for high turbo frequencies and a high number of instructions per clock, which makes them the right place for heavy single-threaded work with minimal latency: an engineering application, a user interface or the serial parts of an AI pipeline.

• Physically larger, with more cache and wider execution resources
• Highest turbo clocks, boosted by Intel Turbo Boost only when extra performance is needed
• Hyper-Threading on 12th to 14th Gen and Core Ultra Series 1, so each P-core runs two software threads
• One thread per P-core on Lunar Lake, Arrow Lake and Panther Lake, where Intel removed Hyper-Threading to save power and area

How many P-cores does an industrial workload need?

The P-core count sets how fast the main application responds; the E-cores carry the rest. These are starting points we use when sizing a system, not fixed rules, and the software itself is the final test.

P-cores are not limited to hybrid chips. Intel’s Xeon server processors have used performance-class cores throughout, and Xeon 6 splits into P-core models for compute-heavy work and E-core models for dense, many-threaded work; our comparison of AMD EPYC and Intel Xeon covers the server side.

Our explainer on Hyper-Threading covers why two threads per core helps some workloads and not others.

An E-core, or efficient core, is the small core in an Intel hybrid processor, designed to deliver the most work per watt rather than the highest single-thread speed. Intel quotes four E-cores fitting in roughly the area of one P-core, so a processor can add many of them without a large rise in size or power. They run one thread each and are built to handle background work continuously, and to join the P-cores on heavily multi-threaded jobs such as rendering video or compiling code.

• Physically smaller, arranged in clusters of four that share an L2 CPU cache
• Lower clocks than P-cores, but strong throughput when many run together
• Ideal for background services, data collection, filtering and transmission
• One software thread per core

E-cores are not slow in absolute terms. The Gracemont E-core in Alder Lake was quoted by Intel as close to a Skylake core of a few years earlier, and later E-cores such as Skymont closed much of the gap to the P-cores. Some processors use only E-cores: our guide to Alder Lake-N covers the four- and eight-core E-core-only parts used in compact fanless systems.

E-core generations

Like P-cores, E-cores change with each generation, and the same designs also power Intel’s E-core-only embedded processors and its E-core server chips.

Skymont was the biggest step. Intel quoted around 38% higher integer and 68% higher floating-point instructions per clock than the Crestmont LP E-core it replaced in Lunar Lake, so an E-core in a current processor does far more than the name suggests.

In an industrial computer, E-cores suit the steady background work that runs alongside the main application: fieldbus and network stacks, data logging, container runtimes, security agents and remote management. Our comparison of the Intel N100 and N150 shows how two E-core-only processors differ in practice.

Low-power E-cores (LP E-cores) arrived with Core Ultra Series 1 (Meteor Lake). They sit on the SoC tile, separate from the main compute tile, so the processor can switch the compute tile off entirely while the LP E-cores keep light tasks running. That cuts idle and light-load power, which matters in battery-powered and always-on devices.

The mix has changed with each series. Meteor Lake has two LP E-cores alongside its P-cores and E-cores. Lunar Lake drops the standard E-cores and pairs four P-cores with four LP E-cores. Arrow Lake-H returns to two LP E-cores, while the Arrow Lake-S desktop parts have none. Panther Lake brings back four LP E-cores alongside up to four P-cores and eight E-cores. Our guide to Intel Core Ultra covers each series in more detail.

The table traces the core mix from the first hybrid desktop and embedded parts to the current generation. Counts are the maximum for each line; individual models have fewer.

For industrial buyers, the practical point is that the newest generations have more cores but fewer threads per P-core. A 14th Gen part with 8 P-cores and 16 E-cores runs 32 threads; an Arrow Lake part with the same cores runs 24. Software that scales with thread count, rather than core count, may need checking. The embedded versions of each generation, with longer availability, are listed in our Intel Raptor Lake, Intel Bartlett Lake and Intel Core Ultra ranges, and our guide to CPU generations and embedded processors sets them in Intel’s roadmap.

Intel’s specification pages list the P-cores, E-cores and LP E-cores separately, alongside the total core and thread counts. Datasheets and manufacturers often shorten this to notation such as “12-core (8+4)” or “8P + 4E”, where the first figure is the total and the brackets give the split, or “8C4c”, where a capital C marks a P-core and a lower-case c an E-core.

On a running Windows 11 system, Task Manager shows the logical processors on its Performance tab, and the Details tab can show which core a process is using; Linux tools such as lscpu list the core types on recent kernels. The thread count is a quick check of the split on 12th to 14th Gen parts, because each P-core runs two threads and each E-core one.

The worked rule for older parts is threads = (2 × P-cores) + E-cores + LP E-cores: the i9-12900TE gives (2 × 8) + 8 = 24. From Lunar Lake onwards, threads simply equal cores.

Intel Thread Director is hardware inside hybrid processors that monitors what each thread is doing, such as the mix of instructions it runs, and passes hints to the operating system about which core type would suit it. The operating system’s scheduler makes the final decision. The aim is to keep demanding foreground work on P-cores and move background tasks to E-cores, so that a heavy job such as machine learning or data analysis is not slowed by routine processes.

How well this works depends on the operating system.

For that reason we recommend Windows 11 for hybrid processors, typically the long-term servicing edition; our guide to Windows 11 IoT Enterprise LTSC 2024 covers editions and support dates. On embedded Linux, a current kernel matters more than the distribution name.

Automatic scheduling suits most applications, but industrial systems often have tasks that must run at a fixed rate, and two cores of different speeds can make timing less predictable. The usual answers are simple.

We set these options as part of the build and test the timing under load, so the system behaves the same on the bench as in the cabinet.

Usually not. E-cores add parallel throughput and keep power down, and with Windows 11 or a current Linux kernel, most software runs well on a hybrid processor. Disabling them lowers the core count, raises power at light load and often makes multi-threaded work slower.

There are a few cases where it helps: older software that assumes every core is the same speed, per-core licensing where E-cores count against the licence, and some latency-sensitive setups where a uniform set of cores is easier to tune. Even then, pinning tasks to P-cores is often a better answer than switching E-cores off. On the newest processors, removing E-cores also removes most of the multi-threaded performance, so it is rarely worth it.

Intel’s hybrid design suits embedded computing because industrial systems often mix steady, low-intensity tasks with bursts of heavy work. For data collection, filtering and transmission, E-cores run continuously at low power, which helps in remote or distributed sites. P-cores provide the processing for compute-intensive work such as AI inference, computer vision and real-time analytics, in applications from industrial robotics and autonomous systems to predictive maintenance.

Power efficiency and thermal management are critical where a computer must run in a compact, often fanless enclosure. An embedded PC can use E-cores for light or background tasks, keeping power and heat down, with the P-cores called on only when the workload needs them; Turbo Boost raises P-core clocks only for those bursts, within the processor base power the heatsink is designed for. That keeps operation stable under load and makes fanless PCs practical even with 10 or more cores.

The approach mirrors Arm’s big.LITTLE design, which also pairs large and small cores, but Intel’s version keeps full x86 compatibility, so legacy industrial software and current operating systems run unchanged. Our guide to x86 vs Arm compares the two architectures for industrial use.

Not every Intel processor is hybrid. Several lines use E-cores only, which gives a uniform set of cores, simple scheduling and very low power, and they are common in compact fanless computers.

For low-power control and gateway work, an E-core-only processor is often simpler to specify than a hybrid one. Our Intel Processor N and Intel Amston Lake ranges cover these systems.

We configure every hybrid system to its job: BIOS power limits set to the enclosure so the P-cores do not throttle on a hot day, E-cores left on unless the software needs otherwise, and real-time tasks pinned where timing matters. We then image Windows 11 IoT Enterprise or Linux with a kernel that understands the core types, install your software and test it under load before it ships.