Unified Core: Breakthrough Change Is Coming To Intel CPUs

Since 2021 Intel has been using a hybrid CPU architecture with big and little cores (starting with 12th-gen Core CPUs, though it had already been tried a year and a half earlier in the Lakefield SoC). However, the little “E-Cores” derived from former Atom SoC are quite unpopular among some. Now, a radical change is emerging on the horizon—instead of maintaining two completely different core architectures, Intel is moving to a single one.

Unofficial leaks and rumors circulating online—sometimes difficult to distinguish from unfounded speculation—have for some time mentioned that Intel is preparing something called Unified Core. As the name suggests, this program aims to unify the two distinct core architectures that Intel currently develops.

The big cores (P-Core) are descendants of Intel’s main processor lineage that powered Nehalem, Sandy Bridge, Haswell, Skylake, Tiger Lake, and Rocket Lake processors. Through Core 2 (2006), these cores are also direct descendants of the P6 architecture hailing from the Pentium Pro processor (1995), which is actually the very design that marked the beginning of Intel’s long-standing technological dominance—and that of the x86 platform—in processor market. Until Pentium Pro, RISC processors dominated in performance, and it was assumed that the x86 instruction set would never be able to compete with them.

The E-Cores, by contrast, descend from the power-efficient Silvermont mobile architecture, which debuted in “Bay Trail” Atom chips for tablets in 2013 and later formed the basis of several generations of low-cost and low-power Celeron and Pentium chips, often somewhat hated for the low single-thread performance of these so-called “little cores.”

In hybrid processors, these two core lineages are used side by side, which creates issues such as instruction set support—Intel’s removal AVX-512 support happened exactly because of this, even though the company had introduced the extension it into big cores only a few years earlier.

According to rumors, Unified Core means that Intel will abandon one of these architectures and replace it with the other—or replace both with a completely new one. However, this most likely does not mean that Intel processors will return to a completely non-hybrid concept with homogeneous cores. It is expected that Intel will continue to differentiate between higher-power P-Core cores specialized for high single-thread performance and lower-power E-Cores optimized for maximum energy efficiency and/or maximum performance per unit of die area (area efficiency). But both types of cores will this time share the same architectural lineage, with each generation spawning two versions featuring modified parameters and details.

This would likely resemble how AMD designs hybrid processors with its “classic” and “dense” cores. The advantage is that both versions would likely be developed simultaneously and share technologies and improvements, instead of being handled by entirely separate teams. There should also no longer be issues with significantly different instruction performance characteristics (although absolute performance will likely still differ), nor problems with E-Cores lacking certain instructions supported by the big cores.

Unified Core confirmed

Although information about Unified Core had been circulating for some time, the fresh news now is that Intel has publicly acknowledged the existence of this undertaking. The company explicitly mentions it in a job listing seeking engineers, recently spotted on LinkedIn, looking for an experienced CPU design functional verification engineer to join the “Unified Core team,” specifically in a position within the “Silicon and Platform Engineering” group.

This is likely the first fully explicit confirmation that Intel is working on the unified core. It is probably still several years away, so there is no guarantee the program won’t fail—similarly to what reportedly happened with the “Royal Core” project, which was said to be an attempt to develop an entirely new core architecture under Jim Keller at Intel. Intel never officially confirmed this project, and we know it has since ended—most likely because the design was too ambitious or relied on concepts that proved to be dead ends, failing to produce a result superior to Intel’s existing architectures.

Unified Core is hopefully on a better path and has progressed further, so we may indeed see it materialize. For Intel fans, this could represent a moment where the company turns a new page and once again demonstrates decisive technological progress—perhaps not unlike in 2006, when it replaced the Pentium 4 with the successful Core 2 processors based on the “Conroe” architecture, leading to the term “Conroe moment.” For Intel, that was an event somewhat similar to AMD’s resurgence in 2017 with the Zen architecture.

Unified Core may be based on E-Core, not P-Core

Interestingly, something noteworthy may be taking shape. According to unofficial rumors, development of the unified core is apparently not based on the current big P-Core cores, now represented by Lion Cove (or the slightly modified Cougar Cove in Panther Lake processors), but rather on the smaller E-Cores, currently represented by the Skymont architecture (Darkmont in Panther Lake).

However, it’s unlikely that performance will drop and processors will only carry little cores in the future, you don’t have to fear that. It probably means that Skymont’s successors will be significantly enhanced to reach the performance parameters of big cores—while a separate, lighter, more efficient derivative will serve as the new E-Core. The shift in big-core DNA from “Cove” cores to “Mont” cores may actually come without any big downsides.

There are indications that the architecture of today’s big cores may have reached its limits and that Intel is struggling to evolve it further—Lion Cove, for example, achieved a relatively modest IPC (performance per 1 MHz clock speed) increase over its predecessor, despite undergoing significant upgrades to its internal structures and core “widening.” At the same time, die area has become very large, suggesting a significantly less efficient design compared to, for instance, Apple—or even AMD (which uses the same x86 instruction set, making comparisons fairere, and moreover uses an older TSMC process with lower transistor density).

Skymont cores, on the other hand, demonstrate solid efficiency relative to die area and has been showing significant generational performance increases over the years. Their architecture appears healthier in terms of being easier and more effective to evolve. It’s possible that due to their more recent origin, they carry less “baggage” and problematic legacy and technical debt that complicate further development. This may ultimately mean positive consequences if they take over from today’s P-Core cores and become the foundation for unified cores.

• Read more: Intel’s new P-Core: Lion Cove is the biggest change since Nehalem
• Read more: Skymont architecture analysed: Intel little core outgrows the big?

Has the Israeli team run its course—time for new blood?

It is possible that the relative inefficiency and sluggishness of today’s P-Cove cores, compared to the seemingly stronger development momentum of current E-Cores, is also linked to their respective development teams. The P-Core cores have been developed for several generations by Intel’s team in Israel (IDC). This team gained renown for “saving” Intel with the Conroe architecture during the troubled Pentium 4 era as mentioned. But nothing lasts forever, and it seems that in turn it is now this team’s performance that may be currently declining and dragging Intel down. By contrast, the E-Core and Atom cores were developed by a team in Austin, Texas. It may now be this team that represents the younger, more capable, and more promising blood within Intel. Notably, the location where the Unified Core team position was opened was Austin.

For this reason as well, an architectural reset based on small cores could be exactly what Intel needs. It might ultimately lead to a repeat of the “Conroe moment”—only this time it will ironically be Conroe’s descendants requiring replacement. Which would not be surprising, considering this is going to transpire more than twenty years later. After such a long period, a reset may indeed be inevitable.

We’ll see in two or three years

How well this will ultimately translate into finished processor products remains to be seen. Intel has not yet announced which processors will be the first to be based on the unified architecture, but we know it will not be the upcoming next-generation Nova Lake, and likely not the generation after that either—which will probably still use a milder evolution of Nova Lake cores. Those processors may thus represent the swan song of the original P-Core lineage historically tracing back to Core 2, Sandy Bridge, Skylake, and so on. Processors featuring Unified Core might appear after that sometime around 2029.

Source: Intel, VideoCardz

English translation and edit by Jozef Dudáš

⠀