Apple A20 Pro preview: Performance, design, AI, and more

When the iPhone 18 Pro and iPhone Ultra launch this September, Apple will power them with the newest A-series chip. As in recent years, the next-generation processor will likely be split into two, with a full-performance “Pro” version and a non-Pro model made for lower-cost products. All of the products released this fall are expected to use the Pro version of the chip.

We won’t know for sure what it can do until we get to test it, but we can examine the history and performance of the A-series processors from the last several years, along with the recent rumors and overall industry trends, to anticipate what we might expect from Apple’s new iPhone silicon this fall. Of course, this is all just an educated guess—Apple is famously secretive, and the details about its products are often not fully revealed until they are released. Sometimes, details are kept under wraps even then.

The first 2nm iPhone chip

The A19 Pro was made with TSMC’s advanced 3-nanometer “N3E” process, building on years of refinement since the A17 Pro first brought 3nm to mass-market devices. TSMC is now ramping production of 2nm process technology, and the A20 Pro is expected to be Apple’s first 2nm product.

This new manufacturing process is expected to allow about 15 percent higher transistor density for chips like Apple’s, along with improvements to both performance and power usage, but just how much better performance or lower power you get out of it will depend on the characteristics of the design.

The bottom line is, this new process gives Apple more room for more cores, more cache, and higher-performance core designs while staying within the strict power and thermal budget required for a smartphone.

Also, just as the iPhone Air features a version of the A19 Pro that has one fewer GPU core enabled, the rumored iPhone Ultra and any iPhone Air successor could very well feature a somewhat handicapped version of the A20 chip as well.

Of note: Apple is said to be using a new chip packaging process, called Wafer-Level Multi-Chip Module (WMCM) packaging. It’s a different method of connecting parts of a multiple silicon dies together in a high-speed fashion. For the A20, it’s likely to mean that Apple can place the RAM chips to the side instead of stacking it on top of the main processor, without sacrificing speed. That takes up a little more horizontal space on the mainboard, but it should provide for better cooling of the main A20 and RAM.

CPU performance holds the crown

Apple’s CPU performance improvements have been remarkably steady over time. Both single-core and multi-core performance shows a drumbeat of regular increases that are almost linear year after year. It’s honestly kind of amazing, if you follow CPU design and architecture trends.

Because Apple’s yearly performance gains are so consistent, we feel pretty confident predicting the A20’s performance. There will likely be a 6-core CPU again, with two performance and four efficiency cores. With the debut of the M5, Apple re-branded its performance cores to “Super Cores,” a meaningless marketing move. Don’t be surprised if the A20 Pro features “Super Cores” as well.

An estimated single-core CPU score of around 4,200 would place the A20 Pro in the same league as the M5 (for a single high-performance core), and way above the best Intel and AMD desktop processors or any processor found in any Android phone. When it comes to phone-capable chips, Apple is so far out ahead in single-core performance that it’s hardly worth mentioning.

Multi-core performance is a different story. Apple’s CPU is no slouch, and will probably break 10,000 in the A20 Pro. But some Android products reach this level, and of course most processors aimed at laptops and desktops do. These all feature far more than Apple’s six cores, so naturally, multi-core performance benefits.

The takeaway is this: For the last several years, there has been no processor in any smartphone that delivers overall CPU performance equal to Apple’s, and the A20 Pro will surely retain that performance crown.

GPU performance for AI and more

GPU performance is tricky. The GPU is used for graphics rendering of course, which has its own performance criteria, but it’s also used for parallel computing tasks like running machine learning algorithms. GPU performance is usually highly dependent on memory bandwidth, which is shared with the rest of the chip.

The GPU performance gain for Apple’s A-series chips has been a little irregular. Sometimes, the GPU architecture doesn’t change much; it just runs at a higher clock speed, perhaps with a bit more cache or memory bandwidth. Other times, Apple makes serious changes to the architecture to increase efficiency and add features.

We don’t know what to expect this year, but we can assume Apple sees the GPU as a big priority, given how important its performance is to AI.

Foundry

The compute performance of Apple’s GPU design took a big leap last year thanks to a new architecture aimed at improving that aspect in particular. It probably won’t shoot up as far again this year, but a boost of more than 10 percent seems reasonable.

When it comes to performance in high-end 3D graphics rendering, as in AAA premium games, again we saw a big improvement with the new GPU architecture last year in the A19 Pro but most other years the gains are less dramatic.

Foundry

Over 50 frames per second in the 3DMark Solar Bay Unlimited benchmark would be a good result. That’s over double the real-world graphics performance in just three years.

Foundry

On the less intensive Wild Life Unlimited benchmark, Apple passed the 100 frames per second threshold two years ago, so it’s not crazy to think we could see a result over 150 this time. That would represent about a 65 percent improvement over three years.

GPU performance is one of those areas where Apple doesn’t completely dominate in its chip design, and it will be interesting to see if the company invests more heavily in it over the next few years, allocating more of the chip’s area to the GPU to cram in more cores, bigger caches, or other performance-improving features. The GPU is very important for local AI tasks, and that seems to be a major focus for Apple. Good news for gamers—you benefit, too.

Neural Engine is more important than ever

Apple has had a dedicated NPU (Neural Processing Unit), which it calls the Neural Engine, since the A11 was introduced with the iPhone X. The first Neural Engine could do 600 billion operations per second, while Apple’s latest chips do over 35 trillion. The Neural Engine is sort of like a GPU, but tailored specifically to run the sort of specialized matrix operations that most AI and machine learning models require.

It’s not just about Siri or Apple Intelligence, either. The Neural Engine is used in dozens of areas throughout your iPhone, from taking photos to recording audio without background noise to copying text out of an image.

With the heavy push to big fancy AI models, NPUs are more important than ever, and the their performance doesn’t necessarily scale with the “trillions of operations per second” statistic. Memory bandwidth, cache, and support for special compact data formats all play a big role in NPU performance.

Foundry

Geekbench has a special separate AI benchmark that runs common machine learning and AI tasks on the CPU, GPU, or NPU. When we look at the results when running on the NPU, we see a some big improvements, including a near-doubling in the last three years.

There’s no question that Apple has big NPU plans. Recent claims have said that the NPU takes up a lot more of the chip area on the A20—if that’s true and Apple can keep up with the memory bandwidth for it, our chart above could be very wrong in a very good way, and we could be in for another big jump in NPU performance as we saw between the A17 and A18.

RAM, N2, and more

Beyond the CPU, GPU, and Neural Engine, the A20 Pro package will be paired with other critical components.

First, there’s RAM. This is a tricky one. It’s probably too soon to expect LPDDR6 memory and the big boost in bandwidth and efficiency it will bring, but Apple will probably use high-speed LPDDR5x memory and maybe a wider memory bus. So many critical features are dependent on high memory bandwidth.

AI also depends on having lots of RAM—LLMs are big. The A19 Pro has 12GB of RAM, while the regular A19 only has 8GB, and it already makes a difference: The new and improved dictation and more emotive Siri voice in iOS 27 require devices that have extra RAM. On the other hand, RAM prices have gone up more than 5x in the last year alone, making RAM super expensive.

Apple probably wants to attach 16GB of RAM to the A20 Pro, but we wouldn’t be surprised to see 12GB, just because the market is totally nuts right now, even for Apple.

Then we have the wireless chips. Apple used its own 5G chip (the C1) in the iPhone 16e and followed up with the surprise C1X in the iPhone Air that’s both faster and more energy efficient. Apple’s new iPhones may use the C1X, but we’re actually expecting the C2, which should further improve performance, add mmWave support in some markets, and improve battery life even further. Notably, Apple’s Pro devices have yet to use anything other than a Qualcomm modem, so we’ll have to see if that changes this year.

Apple has its own Wi-Fi and Bluetooth chip, too: the N1. There’s little reason for Apple to move beyond the N1 yet—it already supports Bluetooth 6 and Wi-Fi 7. An N2 is not expected yet, but if Apple announced an N1X that improves power efficiency, we wouldn’t be surprised.