AMD's hybrid APUs are finally official, but they're not moving the needle much
Key Takeaways
- AMD’s hybrid approach to CPUs differs from Intel’s, with their Phoenix 2 APU not shaking things up as much as Intel’s Alder Lake. The real benefit for AMD is in manufacturing, allowing for smaller and cheaper processors.
- Phoenix 2, AMD’s hybrid APU, is similar to its predecessor but with fewer CPU and GPU cores. It is built on the same process and architecture, with slight differences in cache and features.
- AMD’s choice of a single-CCX design for Phoenix 2 improves core-to-core latencies. The ratio of regular Zen cores to dense Zen cores is likely to remain 1:2 for some time, as AMD may not introduce a new CCX design until a few generations later.
It was only recently that AMD finally made the launch of its first hybrid processor, colloquially (but not officially) named Phoenix 2. This APU features two regular Zen 4 cores and four area- and power-efficient Zen 4c cores, for a grand total of six cores. Intel beat AMD to the punch with hybrid architecture, with Lakefield in 2020 as a proof-of-concept and Alder Lake in 2021 as the real deal. Now, AMD has caught up to its rival and will be making hybrid processors for the foreseeable future.
The thing is, AMD’s approach to hybrid CPUs is very different from Intel’s, and on a per-core basis they’re not going to shake things up nearly as much as Alder Lake and Raptor Lake. Zen 4c is almost identical to Zen 4, and while there are advantages to this, ultimately it means swapping out some Zen 4 cores for 4c won’t make a big difference in performance or efficiency. For AMD, the real benefit of hybrid architecture is in manufacturing, and that’s the thing that might open the door for some truly new AMD CPUs.
What AMD’s first hybrid processor looks like
Although AMD’s hybrid APU is a different chip than the original Phoenix APU that launched earlier this year, its official codename is Phoenix. For the sake of avoiding confusion, I’ll be calling this hybrid APU Phoenix 2, which is what the PC enthusiast community named it when it first leaked out earlier this year.
That being said, Phoenix 2 is basically just a smaller Phoenix and isn’t totally brand-new. It has two fewer CPU cores, eight fewer GPU cores, and is physically smaller. It also lacks Ryzen AI capability and has a slightly smaller L2 cache, though that’s only because it has fewer cores. But otherwise, they’re built on the same TSMC 4nm process, use the same architecture, and have the same amount of L3 cache.
Phoenix | Phoenix 2 | |
|---|---|---|
CPU Cores | 8 | 2+4 |
GPU Cores | 12 | 4 |
Cache | 16MB L3 + 8MB L2 | 16MB L3 + 6MB L2 |
Ryzen AI | Yes | No |
Die Size | 178mm2 | 137mm2 |
What’s particularly interesting is that Phoenix 2 is a single-CCX design. In Zen CPUs, the CCX is a group of cores and is the smallest building block, rather than individual cores. While AMD has previously made two-core, four-core, and eight-core CCXs, Phoenix 2 marks the first time AMD has made a six-core CCX, and going with one CCX means better core-to-core latencies. But that’s not just an interesting tidbit, it’s very crucial for the future of hybrid Zen CPUs since AMD doesn’t introduce new CCX designs very often when it comes to core count.
This all means the ratio of normal Zen cores to dense Zen cores is probably going to be 1:2 for a while, since it’s unlikely AMD will replace the six-core CCX until it’s at least a couple of generations old. The upcoming Strix Point APU is rumored to be a 12-core chip, which means two six-core CCXs. It’s highly unlikely that future APUs built with the six-core CCX will offer more than 12 cores, since more CCXs mean worse core-to-core latencies. If AMD wants to change the 1:2 core ratio or offer more cores per CCX, it will have to introduce a new CCX, but that is certainly years down the line.
How Phoenix 2 compares to Intel’s hybrid CPUs
AMD has taken care to note all the differences between its hybrid designs and those of Intel’s. AMD’s hybrid chips will use cores that don’t differ architecturally, have the same IPC, have SMT/Hyperthreading across all cores, and don’t require complex scheduling. Those are all things that Intel’s current Raptor Lake chips struggle with, as the company’s P-cores and E-cores are architecturally different, while Zen 4 and 4c are identical. However, what Intel CPUs give up in those aspects, they gain in others, and that’s just as true for AMD’s hybrid APUs.
The only difference between Zen 4 and 4c in performance and efficiency is that Zen 4 can hit higher clock speeds, and that’s a double-edged sword for AMD. It ultimately means that adding Zen 4c cores into the mix doesn’t really change the performance or efficiency characteristics when comparing Phoenix 2 to a cut-down Phoenix chip. AMD even admits this quite plainly in its presentation about Phoenix 2, and although Phoenix 2 is more efficient than Phoenix at lower TDPs, it’s a very minor difference that AMD could have accomplished with Phoenix just by tweaking the frequency per core.
Source: AMD
By contrast, Intel’s P- and E-cores use different architectures to offer different power and performance profiles, with the former offering high single-threaded performance and the latter great multi-threaded performance in great numbers. The biggest tradeoff AMD is making is relying on a single core architecture to always meet its performance and efficiency needs. If Intel needs greater single-threaded performance in its next CPU, it just needs to focus on redesigning the P-cores and can simply leave the E-cores alone, for instance.
Additionally, Intel’s current generation Gracemont E-cores offer a much smaller footprint and higher performance density, just like Zen 4c against Zen 4. In fact, Gracemont cores are smaller than Zen 4c cores despite being a generation behind node-wise, but of course Gracemont is much slower than Zen 4c.
It’s not as simple as AMD makes it out to be with its hybrid CPU design, and Zen 4c really doesn’t change much when it comes to performance and efficiency. But that’s the thing, Phoenix 2 isn’t really about performance and efficiency, but rather something else.
For AMD, hybrid design is about manufacturing
The key benefit of Phoenix 2 and other hybrid Ryzen APUs will be in manufacturing. Zen 4c’s more compact size means smaller processors, which are obviously cheaper to manufacture than larger ones. AMD obviously wanted to develop a smaller Phoenix APU for lower-end devices, but without Zen 4c it couldn’t have been so small unless it only used four Zen 4 cores, which would have resulted in much worse performance. Hybrid cores allow AMD to offer the same performance for a lower price, or to pocket the difference and make more money.
While this is a benefit Intel also gets with its approach, AMD is definitely investing far fewer resources by keeping things simple. Cost-effectiveness has been AMD’s motif ever since it launched the first Zen CPUs in 2017, and its hybrid APUs continue that tradition. It will be interesting to see if AMD’s approach towards hybrid design proves to be as successful as chiplets, a concept that Intel is now following with processors like Meteor Lake and Ponte Vecchio.
Additionally, we don’t know if AMD is planning on bringing hybrid design to chiplet-based Ryzen CPUs. Theoretically, AMD could combine a standard eight-core Zen chiplet with a 16-core C-type Zen chiplet (which is currently exclusive to the datacenter) and easily create a 24-core CPU, which may be appealing to AMD since desktop CPUs have been stuck at 16 cores since Ryzen 3000. Such a CPU would have a triple-CCX configuration, however, and it’s unclear if it would work well or even work at all. We’ll all have to wait and see.