When Intel launched its 12th-generation Alder Lake chips in late 2021, it did something really unique by using two completely different kinds of cores in the same package. Of course, Intel didn’t invent what it calls “hybrid architecture” since Arm was doing essentially the same thing under what it calls big.LITTLE for years. However, on desktop, this was a big deal since it allowed Intel to achieve a high amount of performance while using less power and area than a non-hybrid CPU would have. AMD, meanwhile, has continued to offer just one architecture per CPU.
But that won’t be the case forever, as AMD has already all but confirmed its first hybrid processor is on the horizon. Not only is this a big deal in the technical sense, but it also means AMD is taking notes from Intel for once (a reminder that Intel once mocked AMD’s chiplet strategy and is now making its own chiplets, which are branded as tiles). We don’t know exactly how far AMD will go with its hybrid architecture, but we already have crucial details on what will likely be the company’s first hybrid CPU.
How hybrid architecture can make Ryzen even better
Source: Intel
Although AMD has lots of different CPU products, I’m just focusing on Ryzen for desktops and laptops in this article, mostly because hybrid architecture has traditionally been used for consumer stuff and not much (if anything) else. The points I make here will largely apply to other things like the data center segment, though.
One of the things I commonly see people wonder about is why Intel packs its CPUs with weak E-cores instead of going full P-core. After all, P-cores are way faster than E-cores, so, obviously, Intel is cutting corners, right? In fact, not only are hybrid CPUs like the Core i9-13900K some of the greatest CPUs available today, they wouldn’t even be possible without E-cores, and it comes down to two things: power and area.
Firstly, while P-cores are much faster than E-cores, they also consume more power. For CPUs like the 13900K, less efficiency means less performance since it brushes up against the limit of how much power a CPU can feasibly consume without getting too hot. In addition to efficiency, E-cores are also much smaller than P-cores, and by using a lot of E-cores, Intel can pack more performance into a smaller size. More E-cores can allow multi-threaded programs to scale across multiple cores, while also raking in the benefits of the space savings from using these smaller cores.
By offering different cores optimized for performance and efficiency, hybrid architecture CPUs are able to sidestep a fundamental design conundrum that exists in traditional CPUs. In order to boost single-threaded performance, you need to make cores individually beefier, but this often results in inefficient power consumption and area usage. For better multi-threaded performance, however, you need lots of cores, but power and area inefficiency makes that harder to achieve. By offering the best of both worlds, hybrid architecture sidesteps this core design dilemma.
What a hybrid AMD CPU might look like
Source: AMD
Hybrid architecture has arguably made Intel’s best CPUs, and its hybrid CPUs are designed like every hybrid CPU before it, with all CPU cores sharing the same silicon (much like how many CPUs often incorporate integrated graphics alongside CPU cores). However, the possibilities with AMD are much different because the company also uses chiplets in addition to traditional, monolithic designs. Even though we already know a lot about AMD’s first hybrid chip, there are many more possibilities to consider.
Thankfully, we don’t need to speculate about architecture here because AMD already has big (performance) cores and little (efficiency) cores. Regular Zen cores like Zen 4 would be the big cores, while the brand-new power and area efficiency optimized ‘c’ variant cores, such as Zen 4c, would be the little ones. Although Zen 4c is first debuting as a cloud-optimized server CPU thanks to its ability to put 128 cores on a single CPU, I do wonder if AMD always intended to use it for hybrid architecture or if this is a new plan. By contrast, Intel’s first E-core server CPU has yet to come out.
By offering different cores optimized for performance and efficiency, hybrid architecture CPUs are able to sidestep a fundamental design conundrum that exists in traditional CPUs.
We already know some of the key details of AMD’s Phoenix 2 APU, which is the first hybrid chip the company will launch. We know it’s a six-core APU, and we can reasonably assume that it has two Zen 4 cores and four Zen 4c cores, and the end result is that Phoenix 2 is significantly smaller than Phoenix. However, it’s also significantly cut down compared to the regular Phoenix APU in other places; it doesn’t have Ryzen AI capabilities, and its integrated graphics are limited to four cores, which is a third of the iGPU in Phoenix. So, Zen 4c isn’t the only thing that makes Phoenix 2 smaller.
While Phoenix 2 is being manufactured and may even be in laptops you can buy right now, there’s a catch. The quad-core Ryzen 3 7440U will seemingly use both Phoenix and Phoenix 2 chips, and since AMD obviously wants this chip to perform consistently, that means the 7440U may not take full advantage of the hybrid architecture in Phoenix 2. The 7440U might even only use the Zen 4c cores, but we don’t know this for certain yet. The Ryzen 5 7540U could also use Phoenix 2 (though AMD confirmed this isn’t happening yet), but it won’t take full advantage of the hybrid design either.
Additionally, it’s unclear how beneficial Zen 4c cores will be for mobile. While AMD has said its Zen 4c data center CPUs are more efficient than its regular Zen 4 processors, the company didn’t disclose if Zen 4c is more efficient at the same clock speed or if it’s more efficient because it’s clocked lower. If Zen 4 is just as efficient as Zen 4c at the same frequency, then only its density is a significant advantage. That being said, we’ll probably know in the near future how good Phoenix 2 is once it’s finally launched in earnest.
One problem AMD is running into on desktops is that it can only put two CPU chiplets (also called a Core Complex Die or a CCD) in a mainstream CPU, and that’s left Ryzen stuck at 16 cores since 2019. Getting a higher core count requires a brand-new design which would be expensive and a major headache; obviously, increasing the number of CCDs on the CPU isn’t possible since AM5 Ryzen CPUs just don’t have the room. However, Zen 4c CCDs have 16 cores rather than the 8 on Zen 4 CCDs, and using one of each would allow AMD to hit the 24-core mark without issue.
AMD could also design a new chiplet that contains both Zen and Zen c-variant cores, making it pretty similar to Intel’s hybrid CPUs. However, I don’t think AMD will do this, primarily because it doesn’t like designing new chips unless they would have broad use cases, and these hybrid chiplets would probably only be used for Ryzen. Plus, for technical reasons, each chiplet would likely come with eight Zen cores and eight Zen c-type cores, when ideally, you’d have more Zen c-variant cores than regular ones. AMD could do some architectural modifications to change that, but again, AMD hates spending money frivolously.
Regardless, if AMD chooses to bring its compact c-type cores to the desktop, then we’re probably in for some much, much higher core counts than we’ve ever seen before. Chiplets made the first mainstream 16-core CPU possible with AMD’s Ryzen 9 3950X, and hybrid architecture in Intel’s Raptor Lake brought us the first 24-core processor for the mainstream. With chiplets and hybrid architecture combined, we could easily see a 40-core CPU if AMD combines an 8-core chiplet using regular Zen cores with a 32-core chiplet using c-variant cores.
For AMD, hybrid architecture is natural and perhaps even necessary
The proposed death of Moore’s Law may have profound consequences for AMD and how it designs CPUs. Chiplets are a way to get around the increasing cost of manufacturing processors as well as the declining improvements each new process brings. TSMC’s 3nm process node, which AMD will be using for Zen 5, is particularly poor as it provides, at best, a tiny increase in cache density in addition to a relatively poor gain in analog density (which is what makes cores smaller). For an innovative company like AMD, incorporating hybrid architecture seems like the natural way forward.
Phoenix 2 will be AMD’s first hybrid chip, but it could just be the beginning. AMD is clearly starting off small here with a chip that won’t be exclusively used for hybrid processors, but in the coming generations, I don’t doubt that AMD will try to squeeze every advantage it can out of hybrid architecture. It worked out really well for Intel, so maybe we’ll see hybrid designs power some of AMD’s best CPUs in the future.