Apple’s A16 chip doesn’t live up to its ‘Pro’ price or expectations | #ios | #apple | #iossecurity | #hacking | #aihp
Each year, Apple ushers in a new A-series system-on-chip to pair with the latest iPhones. This year, the company has taken an entirely new path. The new A16 processor is reserved only for the “Pro” model iPhones, while the standard iPhone 14 models get the same A15 we were introduced to last year (the 5-core GPU version that was found in the iPhone 13 Pro models).
The this-year and last-year split is not the only unique thing. The A16 is, more than usual, a relatively minor evolution on the previous SoC. There are a few changes to be sure, but the average user is unlikely to notice them. The differences between A15 and A16 appear to be relatively mild compared to Apple’s typical yearly cadence.
Earlier this year, I made some predictions about the A16 that were led by assumptions that certainly did not all pan out. While a few things were true, the performance uplift in the A16 is about half of what I predicted, and there are fewer major technological improvements visible. Here’s what’s new in the A16, and what you can expect from Apple’s first “Pro-only” A series chip.
What’s changed from A15 Bionic
At first blush, the A16 seems architecturally similar to the A15. There are two high-performance CPU cores and four high-efficiency cores, five GPU cores, and 16 Neural Engine cores to run machine learning and AI algorithms. Just like the A15.
The chip is manufactured on a new “4 nanometer” process from TSMC, according to Apple, making it the first such processor in a smartphone. According to Nikei Asia’s teardown, this is a pricey change. The site found that the chip costs “over 2.4 times more than the A15 version used in the iPhone 13 Pro Max released last year.” The site reports that Apple is paying $110 for each chip, which has raised the iPhone’s production costs “to an all-time high.” To its credit, Apple hasn’t increased the price of the iPhone 14 Pro models over last year.
It’s worth noting, however, that TSMC’s “N4” process is not a 4nm process in the truest sense with TSMC itself even calling it “an enhanced version of N5 technology.” While it’s a more advanced process than earlier A series processors, it is not a real next-generation silicon manufacturing process; you’ll have to wait for the 3nm process next year for such a thing.
Dominik Tomaszewski / Foundry
The transistor count has gone up a few percent to 16 billion (from 15 billion), and it’s likely that most of that higher budget is spent in the new display engine (which manages the iPhone 14 Pro’s display down to 1Hz in always-on display mode and can crank it up to 2,000 nits in bright sunlight), memory controller, and image signal processor.
As for the more general-purpose parts of the processor, they seem to be only barely changed. The high-performance CPU cores carry the code name “Everest” and can clock up to 3.46 GHz, a roughly 7% boost over the A15’s maximum of 3.24 GHz for its “Avalanche” cores. The high-efficiency cores are code-named “Sawtooth” and clocked up to 2.02 GHz, which is almost the same speed the 2.01 GHz of the A15’s 2.01 GHz “Blizzard” cores. While these cores carry a new name, the architectural changes appear minor at best, as they don’t deliver performance outside the expected uplift from the increase in clock speed.
The Neural Engine is still 16 cores, just as in the A15. Apple says it performs up to 17 trillion operations per second, which is roughly an 8% increase over the 15.8 trillion of the A15. I think it’s likely the same design just clocked a little higher.
Perhaps the most significant change is the switch to LPDDR5 memory, which should provide 50 percent more memory bandwidth than the LPDDR4x memory in the A15. Apple actually made the switch to LPDDR5 in the M1 processor line (on the M1 Pro, Max, and Ultra), which is based on the A14 chip architecture–the only real surprise here is that the company waited this long to do it in their iPhone-bound chips. There may be some very specific circumstances where a task is entirely limited by memory bandwidth on the A15, in which case the A16 should perform a lot better.
So at first blush, we have what appears to be essentially an A15 that clocks higher, with a new display engine and perhaps image signal processor. We’ve read reports that there are new security measures in the processor’s ROM; unsurprising, considering how hard Apple works on both the hardware and software side to make their devices difficult to hack.
Given that the CPU architecture hasn’t changed much, but just runs at a clock speed up to 7 percent higher (and with more memory bandwidth available), we should expect most CPU benchmarks to show performance gains of 10 percent or less.
A quick look at Geekbench 5 numbers shows us that, indeed, maximum single-core CPU performance appears to have gone up by around 8-10 percent over the A15. Multi-core performance fares a little better, but it’s likely that those tests are more easily able to overwhelm the chip’s caches and would therefore get some benefit from the increased memory bandwidth.
The A16 Bionic has five GPU cores, just as the higher-end A15 does, and I don’t believe there have been any architectural changes. But high-end 3D graphics tend to be very demanding on memory bandwidth, and I would expect the switch to LPDDR5 memory to have a significant impact here. I don’t have any real insight into the GPU clock speeds, but it would be reasonable to expect that the cores can clock around 7 percent higher, just as the high performance CPU cores do.
Taking a look at one of the most strenuous 3D graphics benchmarks, 3DMark Wild Life, performance varies from around 7 percent faster on easier modes to around 19 percent in the “Wild Life Extreme Unlimited” test. That’s a good improvement, and in line with what I would expect from a mild clock speed increase and big memory bandwidth boost.
When using the GPU to perform general-purpose computations, as tested in the GeekBench compute score (see above), the performance uplift is in the 7-8 percent range.
A15+ would be a more honest name
There’s no doubt that the A16 is not simply a “binned” version of the A15 (“binning” is when chips that are tested to perform better in manufacturing are separated and sold as a different model). This is a new chip. But there are no major architectural overhauls here that I can see, just minor revisions to improve maximum clock speed and power efficiency. This is less of a leap over last year’s model than we are used to seeing in Apple’s yearly iPhone revamp, a fact that is only underscored by the fact that the standard iPhone 14 models are still using last year’s A15 while still providing important features like Action Mode, Photonic Engine, and 4K Cinematic mode.
Apple didn’t promote any particular feature of the processor as “new” other than the display engine (which is required to manage the iPhone 14 Pro’s always-on display and 1Hz refresh rate), and in fact marketed it most directly against Android phones and the A13, it’s three-year-old flagship. The performance charts just don’t look very impressive with a 7-10 percent performance bump.
To that end, I feel like Apple probably shouldn’t have given this chip the A16 moniker. In most ways that matter, it’s a tuned-up A15. Even the new “4nm” manufacturing process is best described as a modified 5nm process. It’s probably unreasonable to expect groundbreaking advancements every year, with entirely new architectures delivering 20% performance improvements. The occasional “tune-up” year is fine, especially since Apple has such a commanding lead in smartphone performance right now. But the naming should reflect that, and a title like A15+ or A15 Pro feels like a more honest representation of this chip.