Wait, the Minor-Update-ish A16 Bionic SoC is Actually "Surprising"?

And what the heck's "sleeper silicon", anyway?

I'm borrowing a term from the automotive world which you may have heard before - the "sleeper car". In a more benign context not involving plot lines from early Fast and the Furious installments, it's an unassuming car hiding higher-than-expected performance underneath the hood - usually to the gleeful delight of its owner.

As hinted in the title, I'm not talking about the entire A16 Bionic SoC in the newly reviewed, launching on Sep. 16, 2022 iPhone 14 Pro-series. But don't get me wrong. A16 Bionic is an impressive chip overall, especially for being on some type of murky "third-generation" silicon process which evidently amounts to a half-step improvement during the industry's impatient wait for 3-nanometer lithography.

Despite the new 48MP wide camera, always-on-display, and brighter display in HDR and outdoor contexts, the A16 Bionic, according to reviews, is still managing approximately the same battery life as before (perhaps a minor net loss), while:

• scoring mid-single-digit (approx. 5-7%) YOY improvements in the GPU via various graphics benchmarks (possibly driven by an implied change to faster, higher-bandwidth LPDDR5 unified memory, but a win's a win), and

• per Geekbench 5 testing, managing at least 8-9% YOY gains for the performance cores, while claiming to operate at up to 20% lower power.

That's not all of the story, however. Now, I get it. Tech reviewers have to carefully craft a well-presented, differentiated review (written, video, sometimes both!) within a short amount of time before the review embargo lifts and it becomes something of a review free-for-all for readers' limited amounts of attention. I get that it's a ton of sprinting being done by traditional-through-social media, and it sounds intense.

It's quite possible, though, that all of the first-wave reviewers (and I read or watched at least ten of them) missed a big change in a small corner of the SoC complex. One that's happened at least twice in the past. Yes, I'm talking about the sleeper silicon that continues to build towards something ambitious in the distance, even if it goes largely unnoticed in this generation.

"Pro. Beyond (CPU Performance Cores)." Way Beyond.

First, let's meta-analyze the Geekbench 5 CPU scores from four just-published iPhone 14 Pro/Max reviews, all of which are linked below, and all of which compared at least one of their review units against a corresponding iPhone 13 Pro or Pro Max (nice going, reviewers!). Yes, I know, GB5 is a "synthetic benchmark", but as you'll see, it's a fully (ahem) apples-to-apples comparison, pitting this year's iPhone Pro lineup versus last year's.

Sources: CNET (Patrick Holland, 14 Pro/Max review); Input Mag (Raymond Wong, 14 Pro/Max review); Tom's Guide (Jordan Palmer, 14 Pro review and Mark Spoonauer, 14 Pro Max review)

There's two pairs of data points per reviewer, but it's easy enough to compare. The shorter bars are GB5 single-core scores, and predictably enough, all 14 Pro-series review units (blue bars) are a bit faster than the 13 Pro or Pro Max each reviewer compared against (grey bars). With all A16 single-core scores in the high 1800s, they are theoretically at the same peak performance level as an M2 performance core (although battery and thermals will necessarily "constrain" the CPUs).

Looking closer at the GB5 multi-core results, the A16 SoCs powering the four review units sure seem to outperform the A15s by a wider margin than single-core, don't they?

To better visualize this, let's check year-on-year multicore percentage deltas:

The Numbers app makes resizing fonts a challenge, but luckily this is a simple chart to explain.

Where the A16 perf cores managed 8-9%-level gains vs. the A15 (blue bars) in GB5, multi-core scores show gains of 15-17% (green bars) in this comparative sample.

Lost in the understandable, chaotic shuffle of iPhone review (deadline) season is how multi-core gains significantly outperform single-core gains. Of course, if every CPU core gets a 10% speedup, multi-core will have an approximate 10% speedup, assuming minimal processing losses due to a well-tuned performance controller.

Obviously, that didn't happen this year.

The only way multicore performance uplift of almost double the percentage of the performance cores happens is when the other portion of Apple's A16 six-core CPU complex - namely, its four efficiency cores - advance at an even faster rate than the performance cores.

Semi-Educated-Guessing a Rough Estimate of A16 Bionic "e-core" Performance

At this point, further derivations are a matter of division and subtraction.

Assuming nominal multicore processing inefficiencies (because the bigger the multicore performance penalty, the faster each individual core actually is), we have an four-iPhone A16 Bionic average of 1882 for CPU single-core, and 5446 for CPU multi-core. Big numbers for a smartphone chip.

Since there's twin performance CPU cores, that leaves an eyebrow-raising 1682 average multiprocessing score solely attributable to the A16 Bionic's four CPU efficiency cores (or "e-cores", as Apple often calls them). This means that the four e-cores might contribute up to 30% of the formidable A16's CPU total multi-threading power, likely their highest combined share yet. In relative terms, the four e-cores may boost CPU multiprocessing performance by around 45% versus just using the two performance cores.

Assuming further that an average A16 CPU e-core would score about 1/4 of its implied multicore contribution, we end up with an approximate "single e-core" GB5 score of about 420. That score is almost halfway between an average GB5 single-core score for the Apple A8 (from the iPhone 6) and the A9 (from the considerably faster iPhone 6s and first-gen iPhone SE). So if Apple ever wanted to swap the iPad Air 2's three-performance-core A8X with the A16's four e-cores instead, it could perform considerably faster, and with astonishingly better CPU power consumption.

And when comparing against the quad A15 e-cores using the same calculations (which contributed around 1230 multicore points to the total A15 Bionic GB5 score), we arrive at an implied e-core speedup of over 35% - a massively higher year-on-year perf increase compared to the plenty-fast performance cores.

This matters because the CPU e-cores have been on a low-key silicon warpath since the A11 Bionic - when the e-cores both doubled in number from the A10 Fusion's two to four, and had a claimed 70% perf jump per core. For the iPhone 8-series and iPhone X, Apple further enabled the new A11 e-cores to act on their own or as four small multiprocessing turbochargers for the much bigger, higher-performance A11 CPU cores (which eke out a GB5 single-core score of around 900). This is how Apple achieved a claimed 70% year-on-year improvement in multi-threaded workloads versus A10 Bionic despite having "only" 25% faster twin performance cores.

Apple's second "out-of-step" e-core advancement was just last year, and flew almost entirely under the radar. While a few too many tech journopundits suffered news amnesia, claiming Apple newly hid behind vague numbers to obfuscate a lack of silicon progress, it turned out that Apple did indeed advance A15 Bionic just fine on the TSMC N5P process (with the A15's higher-spec ~25% faster 5-core GPU featuring in today's iPhone 14 Regular/Plus line).

Just like this year, tech journopundits mostly ignored GB5 results showing a greater CPU percentage uplift in multi-core than single-core. It was an impressive feat of oversight - because looking more closely, the A15 e-cores may have had a staggering performance leap in excess of 50% vs. their A14 counterparts (and the A14 e-cores already had implied multicore potential slightly greater than an iPad mini 4 with upclocked dual-core A8).

The Future of Apple's "Post-Performance Cores"

It's pretty safe to assume that Apple's e-cores, albeit from a smaller base, are still increasing in raw speed at a faster rate than the performance cores. Which leads to a single question with no clear answer - what will Apple do with all of this e-core power?

For Macs, iPads and iPhones, the e-cores should remain a low-key competitive advantage. While the e-cores in the A16 Bionic, based on reviews, might not be able to moderate overall iPhone power consumption on their own (considering the formidable power demands of GPU cores and ever-brighter, always-on OLED displays), they remain comfortably on track to become considerably more powerful (read: greater capability, higher light-duty efficiency) when the computing industry's very bright 3-nanometer silicon future arrives. We may eventually see Pro-level Apple Silicon Macs go from the current two efficiency cores to four, which should benefit battery life for the 14"/16" MacBook Pros while also adding a slight boost to top-end multicore power. Every extra half-hour of runtime helps.

During the iPhone 14 Pro/Max presentation, Greg Jozwiak made a point of highlighting the "4-nanometer-based" A16 e-core's claimed 1/3 power consumption versus any smartphone competition's efficiency cores (at an admittedly unknown equivalent performance level). Clearly, Apple sees value in entrusting a wider variety of increasingly complex processes to Apple's most efficient CPU cores.

It's in even more "constrained" devices, such as Apple Watch, where silicon based on the A-chip CPU e-core could really shine. For some reason, Apple Watch does not appear to have moved to a 5nm system-in-package - it's still rumored to be based on the second-generation-7nm efficiency cores (as reported by Rene Ritchie) in the Apple A13. Going from a 7nm to a 3nm tech platform will allow all the efficiency and sensor-networking performance that Apple Watch could need for years, which could be especially welcome news for the nascent extreme-use-case Watch Ultra userbase.

For other future battery-constrained devices that either can't operate with a smartphone-class SoC, or can simply rely on an iPhone or Mac to handle heavy CPU lifting, the e-core tech platform could be more than sufficient for headset-type wearables - although the ability of augmented or virtual reality wearables to supplant or even supplement the almighty pocketable communications device still known as "smartphone" is anything but clear at this point.

In Apple's silicon philosophy, performance per watt is everything, even as half-cadence cycles such as "4nm" force all leading-edge chip designers to focus silicon improvements in only a few key areas. At least to me, it's obvious that Apple spent much time, effort and maybe a few hundred million extra transistors on new e-cores that might even give Snapdragon 845 performance cores from late 2017 a bit of a scare. And it will be very interesting to see if Apple eventually graduates these cores into an entirely new class of CPU core - known in the competitive ARM chip world as "mid-cores".

Only Apple knows where its perf/W and CPU roadmap will lead. Meanwhile, even amidst an awkward tech node transition, the Hardware Technologies team confidently forges ahead, daring Qualcomm, Samsung, and others to follow.