Intel's “big and small cores”: just a transitional solution?

As technology continues to advance, Intel has launched a “big and small core” CPU in an attempt to find a balance between performance and power consumption. However, is this innovation just a transitional solution to meet current market challenges?

This article will explore the design concept behind it, analyze the task scheduling challenges faced in practical applications, and look forward to its future development trends, bringing readers a new perspective.

The Origin of “Big and Small Cores”

In October 2021, Intel released the 12th generation Core processor, officially introducing the “big and small core” design, which is officially called “heterogeneous multi-core”.

At that time, Intel's competitors (Apple, AMD) were likely to be able to use TSMC's 5nm process, while Intel could probably only use the Intel 7 process.

You know, TSMC's 5nm process is almost a major version ahead of Intel's 7 process. The CPU manufactured using the 5nm process has stronger performance and is more power-efficient.

As the process bottleneck could not be broken in the short term, Intel decided to “completely revitalize” chip design, and the “big and small core” solution came into being.

For a CPU, we can use PPA for evaluation, namely: Performance, Power, Area. In other words, the design goal of the chip is to achieve higher performance, lower power consumption and smaller area.

Intel's “big core”: maximizes single-thread performance and responsiveness. From a PPA perspective, it prioritizes performance, and power consumption and size are not very important. However, the result is a high performance ceiling, but poor energy efficiency (performance per watt).

Intel's “small cores”: provide scalable multi-threaded performance and efficient background task offloading for modern multitasking. From a PPA perspective, it is to achieve good performance with not very large power consumption and size. The result of this is high energy efficiency, but the performance ceiling will be lower.

Ideally, the fusion of high-performance large cores and high-efficiency small cores can flexibly cope with various task challenges. In this mode, light-load tasks can be assigned to small cores to handle, greatly reducing power consumption; large cores can focus on heavy-load tasks without having to worry about other light tasks. For mobile devices such as laptops, this mode can also reduce heat generation and enhance battery life.

However, the reality is not ideal.

Difficulty in task scheduling

The final performance of the “big and small cores” on the computer depends on many factors. Since the applications on the computer often “deal” directly with the operating system, the task scheduling mechanism of the operating system becomes the most critical link.

Since the CPU adopts the “big and small core” design, it integrates two cores with different performances, which makes the scheduling more complicated than the single core architecture. When scheduling these two cores, their respective characteristics and performance differences need to be taken into account, which increases the complexity of scheduling.

Here we briefly analyze how the world's two major system manufacturers, Apple and Microsoft, solve this problem.

First of all, from the perspective of system scheduling strategy: Apple implemented QoS earlier than Microsoft, and its QoS is more detailed than Microsoft.

Intel's

According to the documents on Apple's official website, Apple developers can set the QoS level of the process, and then the system can schedule it according to the QoS priority.

Intel's

In addition, according to the document log, the document was first updated in July 2014 and last updated in September 2016. In other words,Apple has been working on this “solution” for about 10 years.

During these 10 years, applications and systems have reached a tacit understanding, and it has become a certainty that each program should run on a big core or a small core.

Intel's

Intel's

Microsoft only added basic QoS policy in Windows 10 1709 version, and this QoS policy is relatively rough. For example, the service quality in different window states is only divided into three levels: high, medium, and low.

Therefore, from the perspective of system scheduling strategy, Apple wins.

Then we look at the software ecosystem and applications: Apple’s software suppliers are more obedient and more willing to cooperate with Apple to optimize and adapt the “big and small cores”.

Since many Apple users are accustomed to downloading applications from the Apple Store, and applications need to comply with Apple's relevant regulations if they want to be listed on the Apple Store, it is relatively easy for Apple to adapt the entire ecosystem to the big and small core design.

For Microsoft, its long-term strategy is to maintain strong compatibility. For example, some very old software can still run on new systems. However, these old software definitely do not have QoS policies, and Microsoft has no way to force the original developers of old software to come back and update the software.

Therefore, it is difficult to adapt old software to the “big and small core” solution, and new software does not necessarily have to write QoS policies. After all, the presence of the “Microsoft App Store” is very low, and Microsoft has no good way to force software vendors to adapt to the “big and small core” solution.

Therefore, from the perspective of software ecology and applications, Apple wins again.

So from this perspective, if Intel wants to develop a “big and small core” CPU, the best option is to rely on Apple.

However, Intel no longer has this option. When Intel was working on the “big and small cores”, Apple had already started to use its own CPU, the M1 series. Therefore, the combination of Intel's “big and small cores” CPU + Apple system is basically unlikely. In this case, the consumer market is mainly Windows + Intel's “big and small cores” solution.

Although Microsoft is not very good at “big and small core” adaptation, Intel is not completely flat. Intel has launched a hardware thread scheduler (Thread Director), which helps the operating system schedule threads to the core with the best performance and efficiency by identifying the level of each workload and using its energy and performance core scoring mechanism.

However, there is a description on the official website about the hardware thread scheduler that needs attention:“Provides runtime feedback to the operating system so it can make the best decisions for any workload.”

That is to say,The “hardware thread scheduler” can make suggestions about task scheduling to the operating system, but whether the operating system listens to it is another question.

Therefore, the “hardware thread scheduler” is not a solution that can solve the problem independently, and it still needs to rely on Microsoft's cooperation to a large extent. In addition, Intel also needs a long time to polish its own scheduling algorithm to make this scheduling method work well.

Intel 3 process and Xeon 6

The “big and small cores” originated from backward technology and may eventually end due to technological progress.

Some clues can be seen from the Xeon 6 series processors released by Intel not long ago.

Intel's

Xeon 6 processors are divided into two product series: energy-efficient core processors and performance core processors. The energy-efficient core processors are optimized for high core density and high-performance scalability tasks, while the performance core processors are optimized for high performance required by compute-intensive and AI workloads. The two are compatible in architecture, share software stacks and an open ecosystem of software and hardware suppliers.

In short, the Xeon 6 series processors return to the previous “traditional” design ideas.

The reason behind this is that the Xeon 6 energy-efficient core processor uses the Intel 3 process technology. Compared with the previous process node Intel 4, Intel 3 has achieved about 0.9 times the logic miniaturization and 17% performance per watt improvement, which is higher than the general industry standard.

In addition, Intel has become more proficient in the use of EUV (extreme ultraviolet lithography) technology and has added the use of EUV to more production processes of Intel 3.

Intel 3 also introduced a higher density design library, improved transistor drive current, and optimized the interconnect technology stack by reducing through-hole resistance. The Intel 3 process now used by Intel is no longer significantly behind its peers.

In this way, if Intel returns to the “traditional” mode of “only one core in each processor”, the operating system scheduling will be relatively easier. It will also be easier for users to choose the product line they need according to the usage scenario.

A full return to “traditional” design will not happen overnight

The advent of the Xeon 6 processor has brought a new development idea for consumer-grade “big and small core” CPUs. In the context of the continuous advancement of current technology, it seems that it is not so necessary to stick to the big and small core design.

Return to"Tradition"The design concept can simplify complex task scheduling problems and has indeed become a feasible strategy. However, even with such an idea, it is not easy to implement it in practice.

Although Intel now has the Intel 3 process, it will take a long time for it to be fully popularized due to the high cost and limited production capacity of the new process.

For example, Intel’s latest 14th-generation Core processor released in October 2023 still uses the Intel 7 process, and even the Intel 4 process is not used.

The “big and small core” solution does have advantages on low-power devices, so if Intel wants to abandon the “big and small core” solution, it should start with desktop processors that are not sensitive to power consumption, and then gradually move to the notebook platform.

For example, the Lunar Lake processor for notebook platforms, which will be released in the third quarter of this year, still uses the big-small core design. This choice undoubtedly shows that the big-small core design will still dominate the next few generations of notebook platform CPUs.