A big.LITTLE Problem: AMD Better Hustle to Top Intel, Arm on Hybrid Chip Designs

[ad_1]

Processors for PCs are constantly evolving, and introducing a radical new chip design is always a gamble. But Arm’s hybrid big.LITTLE design philosophy, writ large across the smartphone market, has been one of the most significant success stories of the last decade. It has helped to boost performance and efficiency in mobile devices for years. We observed similar benefits when Intel introduced a parallel design scheme with its “Alder Lake” processors in 2021, one of the largest architectural paradigm shifts the company has ever made.AMD has enjoyed success with its own pioneering processor designs in recent years—its “chiplet” innovations, in particular. But it’s late to adopt a hybrid big.LITTLE chip design of its own. Though AMD has made some slight efforts in this vein, it has yet to deploy a true big.LITTLE product. This could change with AMD’s next generation of processors, but before that happens, let’s discuss why AMD needs a true big.LITTLE design in the first place and what that design could look like.What Is big.LITTLE, and Why Does It Matter?Let’s first talk about exactly what a big.LITTLE-style hybrid processor design even is. In processor designs over the last 50 years, many have their unique advantages. Central processing units (CPUs) are more often than not built to be well-rounded, jack-of-all-trades devices, but as usually happens with jacks-of-all-trades, they end up being a master of none. That is why we see more powerful processors designed specifically for supercomputers and less powerful, more energy-efficient ones designed for low-power devices like smartphones and tablets.Hybrid processor designs buck this trend and work around it by typically using two separate processor architectures, each with its strengths and weaknesses. The strengths of each can make up for the shortcomings in the other, nullifying them in some situations. Let’s look at Arm’s and Intel’s designs briefly to illustrate this.Arm kicked off the big.LITTLE hybrid design concept by combining its Cortex A15 cores with Cortex A7 cores in 2011. As Arm licenses its designs out, we didn’t see products based on this design until 2013 when Mediatek, Qualcomm, and Samsung were among the first to implement it in smartphone SoCs. The Cortex A7 cores were exceedingly energy-efficient but lacking, comparatively, in speed. The Cortex A15 cores drove significantly more performance than the Cortex A7 but had a higher power draw. In practice, the goal was to use the Cortex A7 cores as much as possible to conserve battery life, and this worked well, as you don’t need a ton of performance for simple tasks like checking the time, sending text messages, or even working on email. The Cortex A15 cores would kick in when more performance is needed, like while playing games or heavily multitasking. As a plus, some designs enabled all cores to engage simultaneously when even greater performance was needed.Intel did something similar when introducing its first hybrid design in 2021, code-named Alder Lake. Not all Alder Lake processors implemented a hybrid design, but those that did combined a mixture of high-speed Performance cores (P-cores) with power-saving Efficient cores (E-cores). Still in use today, these aim to work exactly like Arm’s big.LITTLE to conserve juice by using the E-cores when possible, and to gain better performance by using the P-cores (or all of the cores together, when necessary). The P-cores inside of Intel CPUs are a direct continuation of the CPU cores used in previous generations of Intel Core processors. In contrast, the E-cores evolved from Intel’s line of Atom processors. The Atom cores were intended for compact laptops, tablets, and even smartphones, so they worked well in their new role alongside the P-cores.In addition to playing to the strengths of more than one processor architecture, hybrid big.LITTLE processor designs have other advantages, particularly in overall core count and cost. The efficiency cores are typically much smaller than their high-performance counterparts and are made of fewer transistors. This makes them cheaper to produce; they take up less space on a chip wafer, enabling more chips to be cut per wafer, which reduces the price per chip due to economies of scale.The smaller size of these chips has also enabled Intel to increase the overall core count of its processors to gain even faster performance, particularly in heavily multi-threaded applications. This is a crucial detail, as one small core may not be a match for one of its large-core counterparts, but as they are smaller, you typically get more than one small core in the space that a single large core would occupy. As you can see from the schematic below, Intel shows four of its E-cores to be roughly similar in size to one of its P-cores.

(Credit: Intel)

In terms of performance, one P-core would be significantly faster in a single-threaded workload, but the E-cores together would likely outperform it in a heavily threaded workload. Essentially, this comes down to the fact that, by trading off a few P-cores for a greater number of E-cores, you can potentially see better overall performance without pushing up the chip size and significantly increasing production costs. Doing this has proven advantageous for Intel and its recent processor designs, giving it an edge in terms of price on some products and overall core count.AMD’s Current big.(not so)LITTLE ChipsAMD’s first efforts to produce a big.LITTLE core design have differed from Arm’s and Intel’s efforts in an interesting way. While both Arm and Intel used entirely different processor designs, with big cores optimized for high performance and little cores tailored to high efficiency, AMD used roughly the same architecture for both. To accomplish this, AMD reworked its existing Zen 4 processor architecture into the similar new Zen 4c architecture that was introduced with the Epyc 9004 Bergamo series of server processors.Zen 4c CPU cores are reportedly about 35% smaller than their Zen 4 counterparts. This was achieved primarily by reworking parts of the processor to be tighter-packed and by reducing the amount of L3 cache per core to 2MB instead of 4MB. AMD claims that the Zen 4 and Zen 4c architectures should drive the same performance, but I am hesitant to trust this entirely, as that added cache will likely give regular Zen 4 CPU cores an advantage in some situations, such as games. However, this is likely of no concern for server customers buying hardware running on an Epyc processor.

(Credit: AMD)

The Zen 4c CPU cores have since debuted on the mobile and desktop markets. On these processors, we see the Zen 4c CPU cores paired with standard Zen 4 CPU cores and Radeon graphics processors in a single-chip solution. These also have a considerable amount of L3 cache. It varies from chip to chip, but in all cases, it’s more than the 2MB of L3 cache per core as featured on the Epyc processors. This switch back to a larger cache pool for these chips makes sense, as they may be used for gaming and have relatively punchy integrated graphics processors (IGPs) built into them, but, likely, these chips are also not quite as small. You’ll find even less difference between the Zen 4 and Zen 4c cores in these products.As interesting as these new products are, they just aren’t the same as the big.LITTLE designs put forth by Arm and Intel, and AMD’s design decisions have some notable consequences. On the plus side, AMD’s design avoids significant discrepancies in terms of performance from one core to the next, which cannot be said for the others.Intel makes clear that its high-performance P-cores and high-efficiency E-cores do not perform on the same level. The E-cores aren’t intended to perform on the same level as a P-core or one of AMD’s Zen 4 CPU cores. This means that when there’s core-count parity between a modern Intel CPU and an AMD CPU, we would generally expect the AMD CPU to perform faster.

Recommended by Our Editors

Sure enough, this holds up to scrutiny. Looking at the 16-core Intel Core i7-13700K and the 16-core AMD Ryzen 9 7950X, we see clear signs that the 7950X was faster on CPU-centric tasks. The 7950X was beaten in many tests in turn by the Core i9-13900K, however, which is primarily due to the 13900K shipping with eight additional E-cores, bringing its total core count up to 24. This isn’t the best comparison, as the 7950X only has standard Zen 4 cores, but the results would be unlikely to change much if half of the 7950X’s cores were Zen 4c cores.At first glance, this may give you the impression that AMD’s design is superior, but consider why a chip maker implements a big.LITTLE design in the first place. It’s all about performance, cost, and efficiency.With its Epyc processors, AMD could squeeze 16 Zen 4c processor cores into a space slightly larger than what eight Zen 4 processor cores would occupy. While this is impressive, moving to a truly small processor core that’s even tinier, more energy-efficient, and less costly to produce than a Zen 4c core would likely benefit AMD.Could an Old Little Core Return?It’s hard to say for sure if AMD is working on a new architecture with a true big.LITTLE design scheme. While these designs have proven beneficial to Arm and Intel, the tech industry never sits still, and AMD unquestionably has new ideas that it hopes will result in an advantage. However, if AMD were to create a new small core, I see a few promising places from which it could start.For AMD, arguably the most promising starting point for a modern high-efficiency small processor core would be to fall back on its now-old “Jaguar” architecture. AMD originally designed processors based on this setup for tablets and thin-client laptops, but Jaguar cores were also used inside Microsoft’s Xbox One and Sony’s PlayStation 4 game consoles. The Jaguar architecture is too old to serve as a small core in a modern processor, but the design could be modernized with a few design elements from Zen 4 and moved to a new fabrication process.Alternatively, AMD could design a new core from scratch, or further strip down its Zen 4c architecture to create a small one. Knowing which of these options would be best is difficult, as I am not a chip engineer designing processors for a living. But, in the past, I have seen old designs come back in newer products, which leaves the impression that updating the old Jaguar core could be AMD’s easiest path forward for obtaining a small, high-efficiency E-core of its own. Frankly, if AMD wants to remain competitive, given the signs of Qualcomm breaking into PCs (for real, this time!) and Intel’s “Lunar Lake” chips incoming in Q3, it might need to plumb the past. AMD’s next-gen processor design is expected later this year, and soon, we will likely get a look at precisely what AMD has been developing. AMD’s CEO, Dr. Lisa Su, will give a 90-minute keynote at the Computex Taipei trade show on June 3. AMD will likely be teasing its upcoming processors at this event, and we may get our first look at a true AMD hybrid processor at this time. If not, then here’s hoping what AMD developed as an alternative will be exciting and competitive.

Get Our Best Stories!
Sign up for What’s New Now to get our top stories delivered to your inbox every morning.

This newsletter may contain advertising, deals, or affiliate links. Subscribing to a newsletter indicates your consent to our Terms of Use and Privacy Policy. You may unsubscribe from the newsletters at any time.

[ad_2]

We will be happy to hear your thoughts

Leave a reply

Megaclicknshop
Logo
Compare items
  • Total (0)
Compare
0
Shopping cart