The iPhone 14 Pro’s A16 chip: Power and performance beyond a speed boost
July 25, 2022 July 25, 2022
Each year, we take a look at the trajectory of Apple’s A-series silicon (which premieres in iPhones and is often used in iPads) in order to get an idea of how the next chip will perform. By combining what we know of state-of-the-art chip manufacturing and packaging processes and Apple’s past improvements, we can generally glean what’s coming from the next system-on-chip (SoC).
This year, as we look ahead to the A16, making this call is harder than ever. The strongest rumors suggest that the A16 will not move forward to a major new manufacturing process node, and that it will only appear in the iPhone 14 Pro and Pro Max–with the regular iPhone 14 models utilizing the same souped-up A15 in the iPhone 13 Pro models. What’s more, we have to consider the way Apple uses these designs. Prior A-series chips might get a bigger “X” version intended for the iPad Pro, but now that Apple scales the design up to a whole line of M-series chips destined for Macs, including very powerful computers like the Mac Studio, some of those extra cores might make their way into the iPhone processor.
So let’s dive in and attempt to predict the A16, perhaps with a slightly bigger grain of salt than usual. We’re making our best-educated guesses here, but there are more variables at play than in prior years, and we’re often surprised by at least a few details when Apple unveils its new iPhone chips.
A Pro-only chip
Generally speaking, the latest A-series SoC powers the entire new iPhone lineup in the fall. Sometimes the Pro models have a little more RAM, but in broad strokes it’s the same chip. With the iPhone 13 line, Apple changed that a little. The iPhone 13 Pro and Pro Max so have more RAM (6GB compared to 4GB in the iPhone 13 and 13 mini), but it also has five GPU cores where the non-pro models have four. It’s almost certainly the same chip design with one of the cores disabled, a common tactic to improve yields with advanced chip fabrication, but it’s the first time Apple has made this differentiation in iPhone models. (Perhaps confusingly, both versions of the chip are still called just “A15,” and you need to dive into the tech specs to see the difference.)
With the A16, most tipsters are suggesting that Apple plans to include the new chip only in the iPhone 14 Pro and Pro Max. The non-pro iPhone 14 models will allegedly will get the A15, albeit the 6GB, 5-core version found in the iPhone 13 Pro models. It’s unclear yet if Apple will essentially rename that chip “A16,” but it does mean a small but respectable upgrade for the non-Pro models, even though they’re not getting the new SoC.
This is especially important because it could help signal what the A16’s design priorities might be. If it’s only going to show up in iPhones that are considerably more expensive (and rumors suggest the starting prices could rise by $100), it could perhaps afford to be larger and more expensive, with different emphasis. In addition to other features like an always-on display, the iPhone 14 Pro models are said to introduce a new 48MP rear wide camera. That could require a lot more image processing and machine learning power.
A new old manufacturing process
The most believable rumors say the A16 will be manufactured on a 5nm manufacturing process, just as the A15 and A14 were. It’s not often that Apple goes three years in a row on the same process node. Some rumors have suggested that it will be manufacturing with a 4nm process, but TSMC doesn’t actually have one. What it has is N4P, a code name for its third generation of 5nm high-performance manufacturing process. N4P offers and 11 percent performance boost, 22 percent power efficiency improvement, and 6 percent improvement in density over the original 5nm “N5” manufacturing process.
It was the A14 that was made with the N5 process, and the A15 used the second-generation N5P process, which offered some small improvements. In other words, Apple’s not going to get a ton of performance or power efficiency out of a third-generation 5nm manufacturing process. It’s certainly not the leap we expect from TSMC’s 3nm process, which is supposed to be ready this year but probably not in time for Apple to manufacture tens of millions of them for an iPhone 14 Pro launch this fall. When we go to 3nm, Apple will be able to squeeze some 70 percent more transistors in the same space, with considerable power savings.
Despite the fact that this third-gen 5nm process isn’t drastically better than the second-gen one Apple used last year, we think the A16 will have to get substantially bigger. We’re probably looking at a big 18 billion to 20 billion transistors, up from 15 billion in the A15. The bulk of that is likely going to go into the Neural Engine for machine learning, the image signal processor and video encoder and decoders, and some general improvements to CPU performance.
There’s little point to Apple expanding beyond the current core configuration of iPhones. Two high-performance cores and four high-efficiency cores are more than enough for a platform that nearly always runs one app at a time, full screen, with some lightweight background processes and such from other apps. At least, for the time being.
Apple’s probably more interested in improving power efficiency and making CPU cores run faster. The high-performance cores, in particular, are likely an area of focus as the company relies on those to really drive home Mac performance when the CPU core design ends up in M-series chips. The M3 line is likely to have the same CPU core designs found in the A16, only scaled to higher core counts, and perhaps with larger caches.
A rumor from an untested source on Twitter reported recently that the A16 is testing 42 percent faster than the A15. That’s almost certainly an unreasonable expectation. I think Apple will move to LPDDR5 this year, which should improve memory bandwidth, and some other improvements, combined with higher peak clock speeds, might get Apple up to a 15 percent performance improvement.
There’s one exception worth talking about. When ARM was finalizing the ARMv9 instruction set last year, there was some speculation as to whether Apple would be the first to incorporate it in a design. I don’t think it was quite done in time, but Apple likes to be on the bleeding edge here. The ARMv9 instruction set is already available in several core designs licensed from ARM (Cortex-A710, Coretex-A510). Those are going to be used in chips like the Qualcomm Snapdragon 7 Gen 1, which will show up in Android phones late this year.
Apple doesn’t license core designs from ARM; it makes its own designs that are compatible with the ARM instruction set. The A15 supports the ARMv8.5-A instruction set, and the move to ARMv9 could have a positive effect on some very specific areas of performance. In particular, wide SIMD vertex operations (which are usually more of a high-end desktop thing) should be a lot faster and more flexible.
ARM has spoken about performance improvements of 30 percent in the same breath as promoting the ARMv9 instructions, but the company was specifically talking about its own licensed CPU core designs there. There isn’t necessarily anything about the instruction set that would boost performance like that, and Apple already uses its own designs with its own customizations and extensions.
To make a long story short, the A16 might be Apple’s first ARMv9 compatible design and one of the first on the market at all, and that might make some very particular kinds of CPU operations go faster, but overall performance isn’t likely to skyrocket.
Apple continues to invest heavily in GPU performance, despite the fact that iPhones are usually miles ahead of other premium smartphones in this area. Now that the company scales its chips designs up to laptops and desktops, with powerful integrated graphics replacing discrete AMD GPUs, it’s even more important.
Without a huge boost in transistor density from a move to the next major process technology node, Apple probably can’t afford to boost the number of GPU cores a lot. We think the A16 will have either five GPU cores (like the A15’s full configuration) or six, but probably not more. Still, architectural improvements and a shift to LPDDR5 RAM for increased memory bandwidth will help increase graphics performance.
We think it’s reasonable to expect a 25 percent to 30 percent improvement in GPU performance, roughly in line with the last several A-series processors. You’ll especially see this in benchmarks and tests that are currently limited by memory bandwidth.
Apple will probably have to include hardware to accelerate ray tracing in a future GPU. It is quickly becoming standard faire in laptops and desktops, and the M series chips derive their architectures from Apple’s A series. (Or more accurately, they’re all designed as a family of products sharing major architectural elements.) Even the major mobile chip designs from Qualcomm and ARM have ray tracing on their roadmaps.
However, I wouldn’t expect this to be the year it happens. High-performance ray tracing acceleration adds a lot to transistor count, and without a big increase in transistor density Apple will probably conclude that any increase in chip size would be better spent elsewhere. When Apple shifts to TSMC’s 3nm process, ray tracing graphics hardware could be on the table.
A heavy investment in ML and imaging
One of the most persistent rumors regarding the iPhone 14 Pro is that it will feature a new 48MP wide rear camera. In low-light situations it would “bin” the light values of neighboring pixels to produce a 12MP pixel, but in brighter light you could get an image with four times the resolution of today’s iPhones. This also opens the door to 8K video, impressive digital zoom capabilities, and more.
But it asks a lot of your phone’s processor. You can’t just slap in a 48MP sensor and call it a day. You have to have wider and faster data paths to the image signal processor. That image processor has to be able to handle four times the pixels (or else take four times as long to process the image, unlikely given Apple’s priority on camera responsiveness). Modern smartphones produce a single image by taking a of exposures and processing them into a single image across many steps, utilizing all kinds of photo science and machine learning algorithms along the way. Some of these steps could be processed at a lower resolution, but some would need to be done on the full 48MP image to get a quality result.
In other words, this new camera will likely demand a much more powerful image signal processor and Neural Engine (the dedicated hardware for AI and machine learning tasks). I’m sure Apple would love to improve the Cinematic Mode introduced with the iPhone 13 Pro. It’s currently limited to 1080p, for example, but a boost to 4K would require more image processing power. So too would more natural artificial bokeh or tracking of multiple subjects. All good reasons for Apple to seriously beef up the image processor and Neural Engine.
Of course, video encoding and decoding is a priority, too. Apple’s latest video processors handle both HEVC and ProRes, but not the newer AV1 codec. I have long suspected that Apple would be among the first to really popularize that codec, the same way it did with HEIC images and HEVC video. It will deliver equivalent video quality with much smaller file sizes and bandwidth requirements. Will the iPhone 14 Pro be the first phone to offer native default AV1 video support in both hardware and software?
LPDDR5 RAM (for real this time)
Last year, I thought Apple would ditch LPDDR4x memory in favor of LPDDR5. It didn’t. But the M1 Pro and M1 Max in the MacBook Pro, and M1 Ultra in the Mac Studio do, as does the new M2 in the MacBook Air. Is this the year LPDDR5 finally comes to iPhone? Yes! Maybe! Supply chain analysts like Ming-Chi Kuo say that Apple has ordered LPDDR5 for the iPhone 14 Pro models, and it does seem like it’s time. Premium Android phones have used it for some time now, and Apple’s latest M-series chips do, too.
The move from LPDDR4x to LPDDR5 will improve memory bandwidth by as much as 50 percent, and should have a positive impact on energy efficiency as well. This in itself doesn’t make any one task faster, it just gives more breathing room for high-bandwidth operations like 3D graphics and image/video processing.
An upgrade to the Snapdragon X65 modem
The cellular modem isn’t really part of the A16 chip, but it’s a tightly integrated part of the whole platform and critical to the iPhone, so it’s worth mentioning. Apple has been hard at work on its own 5G modem for years, but the latest rumors show it’s not ready for deployment yet.
We expect the iPhone 14 Pro (and possibly the regular iPhone 14) to incorporate the Snapdragon X65 modem. The chief improvements, according to Qualcomm, are better power efficiency and reliability. It’s got a higher theorhetical maximum speed (10Gbps), but you’re unlikely to see that from any real-world 5G network. Rather, the modem is said to be better at reliably locking on to signals, which should allow your phone to get better real-world throughput and use less battery in less-than-ideal situations.
The rumors about when we can expect an Apple-made cellular modem in our iPhones and iPads are mixed. Some say as early as 2023, while others suspect the project has had major setbacks and we’re still several years away. Whatever the answer, this is almost certainly not Apple’s year.