Yearly, Apple introduces a brand new A-series processor with its iPhone upgrades within the fall. We don’t count on something totally different this yr. In truth, the soar to a brand new manufacturing course of know-how—TSMC’s 3nm course of—offers the A17 a chance to be probably the most important leap in efficiency and options in a number of years.

By wanting on the previous decade of in-house Apple-designed A-series chips, along with what we all know concerning the manufacturing know-how out there and the corporate’s course and objectives, we are able to piece collectively a reasonably good educated guess about what to anticipate from the A17.

Only for iPhone 15 Professional (or Extremely)

Final yr, the A16 was unique to the iPhone 14 Professional, whereas the usual iPhone 14 used the A15. We count on a repeat of that sample this yr, the place the brand new A17 is unique to the iPhone 15 Professional and iPhone 15 Professional Max (or iPhone 15 Extremely, if the rumors are true) and the usual iPhone 15 will get the A16 used within the iPhone 14 Professional fashions.

Is that this the way in which ahead for Apple for the foreseeable future? With smartphone options and high quality reaching one thing of a plateau and Apple’s cellphone chips operating circles round most Android telephones, there’s little motive to alter. For Apple, it reduces prices and helps create an even bigger differentiator between common and Professional fashions, which tends to push clients towards the dearer iPhone.

Apple’s first 3nm chip

The A14, A15, and A16 have been all made utilizing a 5nm manufacturing course of from TSMC. Granted, that course of has advanced over time, producing chips which are denser and have improved energy effectivity, however there’s nothing fairly just like the leap to the subsequent main course of node. And that’s what we’re virtually definitely going to get with the A17—the primary large-scale client chip made with TSMC’s 3nm course of.

I wrote at size concerning the benefit Apple will take pleasure in with a 3nm course of, and the massive one is extra density–whereas the A16 was about 16 billion transistors, we are able to count on nicely over 20 billion for the A17, maybe as excessive as about 24 billion.

The 3nm course of affords extra energy effectivity, with a comparable chip at a comparable pace, however Apple isn’t going to make a comparable chip at a comparable pace. Most energy draw will probably be restricted by battery measurement, thermal dissipation, and different components, and I don’t suppose we are able to count on an enormous shift in battery life from the transfer to 3nm alone. At the least, not for energetic use below full energy–not solely will the chip probably devour practically as a lot energy in that situation, however the show and radios additionally contribute a lot to the ability drain.

The place we may even see some enchancment is in standby mode, which can get noticeably higher with the transfer to the 3nm course of.

Foundry

CPU efficiency and options

ARM launched its v9 structure in 2021, and we thought the A16 is perhaps Apple’s first chip to assist the brand new v9 instruction set. As an alternative, it helps ARM v8.6 with loads of Apple’s personal extensions. This yr, with the next transistor funds, it appears ARM v9 assist is probably going.

What benefits do the ARM v9 instruction set and structure present? Apple designs its personal CPU cores, and most of the efficiency advantages promised by the v9 structure are already realized in Apple’s designs and ARM extensions. Certainly, the Snapdragon 8 Gen 1 was one of many first high-end smartphone CPUs with ARM’s Cortex-X2 core with assist for ARM v9, and Apple’s A15 outperformed it by a large margin.

You’ll see plenty of claims that ARM v9 affords a 30 p.c efficiency enchancment over ARM v8, however that’s for ARM’s personal core designs, and doesn’t have in mind the usage of customized extensions. Apple’s in an entire different league right here–we in all probability received’t see 30 p.c sooner CPU efficiency from the A17.

Apple’s new CPU cores for the A17 will virtually definitely be sooner, however not essentially due to the shift to ARM v9. CPU core efficiency is influenced by the instruction set, department prediction, instruction decode, execution items, cache construction and measurement, clock speeds, and plenty of different components.

So far as normal core depend goes, there doesn’t appear to be a lot motive for Apple to maneuver past the 4 effectivity cores and a pair of efficiency cores which have been with us because the A11 Bionic. We simply count on 15 p.c extra efficiency out of them.

Foundry

Merely projecting the final a number of years of CPU efficiency enchancment, we are able to in all probability count on a Geekbench 5 single-core rating between 2,100 and a pair of,200, and a multi-core rating of simply over 6,000. Geekbench 6 was simply launched and we don’t have years of benchmark information to create an correct projection, however a single-core rating of over 2,800 and a multi-core rating of seven,300 or extra appears justifiable. A latest leak that claims a single-core rating of 3019 and a multi-core rating of seven,860 isn’t out of the realm of chance—particularly since we noticed a larger-than-expected leap when Apple moved from a 7nm course of to a 5nm one with the A14—however numbers this early are probably fully made up.

Foundry

In case you’re questioning how that stacks as much as different processors, that may give the A17 a single-core rating much like the newest Ryzen high-end desktop CPUs and Thirteenth-gen Core i7 Intel processors, however a far decrease multi-core rating (which is smart, as we’re speaking about solely two high-performance cores versus 12 or extra in these desktop processors). The A16 already soundly beats Android telephones with the top-tier Snapdragon 8 Gen 2, and the A17 ought to solely widen the hole.

If we’ve realized something over time, it’s simply how regular the efficiency enchancment is for Apple’s CPUs. Single- and multi-core efficiency goes up at a virtually straight line, regardless of which years have large architectural adjustments or manufacturing course of leaps. It’s very affordable to count on the same enchancment this yr.

GPU efficiency and options

The GPU is one space the place the A17 might doubtlessly be very fascinating. Apple has been rising GPU efficiency by a median of round 20 p.c or so with every new A-series chip, although it may be anyplace from 15 p.c to 30 p.c. What hasn’t modified an entire lot is the general function set of the GPU. It’s getting sooner, and there have been some minor new options like variable rasterization charges and SIMD enhancements for GPU compute, however Apple is years behind desktop GPUs in necessary options like ray tracing acceleration.

A sketchy rumor mentioned that the A16 was meant to have a serious GPU structure but it surely wasn’t prepared in time, so it received the identical GPU because the A15 (however extra reminiscence bandwidth improves efficiency). I don’t know if that’s true, however Apple hasn’t up to date its Steel function set tables for builders to incorporate the A16, which is telling.

I feel it’s probably that Apple has an up to date GPU structure able to go for A17. Options like ray tracing acceleration will not be important for iPhone, however this GPU design will discover its approach into future M-series Mac processors, the place the dearth of superior GPU options like ray tracing acceleration put them far behind state-of-the-art.

We will additionally count on improved efficiency in present 3D video games and purposes that use the GPU for computation. When the structure adjustments, the speedup tends to be variable–some issues get quite a bit sooner, and different issues not a lot.

Foundry

Staying the course with a roughly 20 p.c speedup produces a Geekbench 5 GPU compute rating of over 18,000. Bear in mind, that benchmark measures the computational efficiency of the GPU, not its capacity to render 3D graphics.

Foundry

For 3D graphics, a 20 p.c speedup would consequence within the fashionable 3DMark Wild Life Limitless take a look at operating at round 88fps versus 74fps with the A16. At the moment, the Snapdragon 8 Gen 2 is quicker than the A16 on this take a look at (and different 3D graphics checks), however this might put Apple barely forward.

Relative to CPU enhancements, GPU efficiency is extra diversified from yr to yr. The advance from A15 to A16 was muted, owing principally to slight clock pace enhancements and improved reminiscence bandwidth. This yr, if we get each a brand new GPU structure and a serious manufacturing course of enchancment, the leap may very well be a lot bigger.

The media engine is usually loosely related to the GPU, and so that is pretty much as good a time as any to say we are going to as soon as once more hope for {hardware} to speed up encode and decode of the AV1 format. It’s in a lot of the new era of PC GPUs, in any case. We count on Apple to proceed to spend money on the efficiency and energy effectivity of its encoders for H.264, HEVC, and ProRes codecs.

A continued give attention to ML and AI

Apple is large into machine studying and AI. Whereas it doesn’t seem like pushing as onerous as its rivals within the generative AI race led by tasks like ChatGPT, Midjourney, and Secure Diffusion, the corporate has used AI and machine studying all through its working system and apps for years. New options like the flexibility to pick textual content in images are carried out on a regular basis, and Apple dedicates plenty of space to its Neural Engine (the accelerator for machine studying duties).

Within the A16, Apple didn’t appear to alter the Neural Engine a lot in any respect. It’s nonetheless 16 cores, and at 17 trillion operations per second is barely 8% sooner than the A15’s Neural Engine. That’s a sufficiently small leap to imagine it merely runs at the next clock pace. Frankly, we anticipated rather more.

With the 3nm course of making a a lot greater transistor funds attainable, the Neural Engine will in all probability take a big leap. It may very well be simply extra cores, large design adjustments to the way in which the cores function, or each. I’d be stunned if it doesn’t ship over 20 trillion operations per second, although there may be some argument over whether or not “trillions of operations per second” is actually the easiest way to measure efficiency.

Sooner LPDDR5x RAM

With the A16, Apple boosted the RAM to LPDDR5 (from LPDDR4x within the A15). Competing top-tier chips just like the Snapdragon 8 Gen 2 use LPDDR5x, which delivers about 33 p.c extra bandwidth and decrease reminiscence latency whereas utilizing the identical energy.

Extra reminiscence bandwidth is nice for every part, particularly when it doesn’t use extra energy. The obvious beneficiary will probably be high-end 3D video games, which stress each CPU and GPU in ways in which push the boundaries of reminiscence bandwidth.

Apple’s not at all times the quickest to leap on new reminiscence requirements, but it surely does prioritize reminiscence bandwidth and large caches and appears to acknowledge the advantage of shifting issues round sooner as a substitute of simply processing issues sooner. I give it 50/50 odds that we’ll see LPDDR5x RAM within the A17.

Nonetheless a Snapdragon modem (for now)

Apple is anticipated to begin utilizing its personal 5G modems beginning subsequent yr, in all probability with the iPhone SE within the spring and, if issues go nicely, later within the iPhone 16 line within the fall.

The Snapdragon X70 might be the modem Apple will use for the iPhone 15 line (not less than the Professional fashions). Many of the X70’s top-line options sound mainly the identical because the X65 that you just discover within the iPhone 14 Professional, but it surely has a small built-in AI processor that always screens and optimizes connection states, which is meant to end in extra steady and optimum connections. That it’s supposed to enhance real-world pace and increase battery life as nicely.