Apple's A15 Bionic is faster against the "competition," less so against its predecessor

nanoguy

Posts: 1,355   +27
Staff member
In context: For years, when Apple revealed a new A-series chipset, it would call it "the fastest we've ever made" and proceed to compare it with its predecessor. Every new SoC would see incredible performance gains that would look good during keynotes and manifest through both benchmarks and a lot of satisfied iPhone users. This year, however, the equation has changed to "frankly, the competition is still playing catch up with our chips."

On the surface, the A15 Bionic in the new iPhones is an impressive bit of kit, with over 15 billion transistors etched using TSMC's 5 nm process node. The CPU configuration has remained largely the same as that inside the A14 chipset, with two new high-performance FireStorm CPU cores paired with four IceStorm energy-efficient cores. Apple didn't offer a performance comparison with the A14 and instead chose to highlight the fact that its new A15 chipset provides a 50 percent faster CPU performance when compared to the competition.

The new iPhone 13 and iPhone 13 Pro are both powered by an A15 chipset, but they have a small yet important difference: the GPU in the iPhone 13 Pro has one extra core for a total of five cores and support for a variable refresh rate. That said, the iPhone 13 is no slouch either, with Apple claiming 30 percent faster graphics performance—when compared to the competition. We're starting to see a pattern here.

Apple has kept the same Neural Engine with a 16-core configuration, but now it can squeeze more performance out of it—15.8 trillion operations per second, to be precise. The company again avoided a comparison with the NPU in the A14, which is capable of 11 trillion operations per second in machine learning tasks.

Other details are not clear at this point. For example, Apple said it integrated new video encoding and decoding capabilities into the A15 design but didn't mention what those were, exactly. Many speculate that since the A14 already supports AV1 decoding, the A15 may well be able to accelerate AV1 encoding tasks.

Apple won't say, but future A-series chipsets will bring smaller improvements that no longer warrant a comparison with their predecessors. In the case of the M1 SoC, it made sense for the Cupertino giant to make performance claims against Intel processors in the same class. With the A15 Bionic, the company is now drawing comparisons to flagship chipsets from competitors like Qualcomm, Samsung, and MediaTek.

There are three reasons for this behavior. The first is that A14 and A15 chipsets have a large enough lead over the competition that Apple can easily make that a selling point instead of the usual "our fastest yet" claim. This is why the company's tune this year was, "frankly, the competition is still playing catch up to our chips. Not just from last year, but from two years ago."

The second reason is related to diminishing performance gains expected with upcoming process nodes. For reference, TSMC says its N4 node will offer a lot more logic density, but the overall improvements will be a balancing act between up to a 15 percent speed gain and up to a 30 percent gain in energy efficiency. That means the A16 Bionic will probably be a more modest upgrade over the A15 or even the A14 Bionic.

Finally, Apple has been dealing with an exodus of talent that may have delayed plans for a more significant architecture upgrade. The most notable is the departure of chip designer Gerard Williams, who left Apple to co-found Nuvia with two Apple veterans. He was followed by several Apple engineers, and earlier this year, his company was snapped by Qualcomm for $1.4 billion.

As you'd expect, Apple wasn't too happy about this development and dragged Williams into court for allegedly using the former's trade secrets to boost his new venture. But the company has other things to worry about—its senior engineers keep leaving to form their own companies. As reported by Dylan Patel over at SemiAnalysis, some of them have joined a mysterious new startup called Rivos, which is working on a high-performance RISC-V CPU design, likely the first of its kind.

What this amounts to in the end is more competition in the custom silicon race, which is bound to gradually erode Apple's leadership in performance per watt. Things are about to get interesting, so watch this space.

Permalink to story.

 
If Apple’s generational improvements are going to be modest at best going forward, then its fanfare transition away from x86 might not look so good in hindsight. Especially given that Intel is promising some pretty solid performance increases with Alder Lake, with Raptor Lake to follow, and then Meteor Lake. And of course Zen4 and Zen5 ought to be pretty solid updates as well. It would be quite laughable if the incumbent x86 vendors deliver future solutions that are superior in every way to Apple Silicon… making everyone wonder why did Apple leave x86 again?

Of course, only time will tell and we have to wait for final silicon year after year to make any conclusive determinations. Will continue to monitor.
 
I think its pretty obvious the biggest cause is the losing talent point. Money can buy you a lot but if you cant keep talent its a moot point anyway. May apple be experiencing what IBM did?

Comparing against the competition is a bad idea. Intel tried that and it spectacularly backfired. What it tells consumers is "this really isnt an upgrade, so look at these other numbers!".
 
Even though Apple still has the first and biggest sit at the table of TSMC they might still be thinking "Even with our own silicon design, we're still highly dependent on another company that's currently all but extorting more money out of everybody but us and might not even deliver on their performance promises"

I think that with all their incomprehensible quantity of money, their next logical step must be for Apple to start thinking if not their own fabs outright, someone they own and control exclusively. I know it seems implausible to impossible but again, it would take impossible amounts of investment and if anyone has that kind of money, it's Apple. In fact it's *only* Apple and almost nobody else.
 
Even though Apple still has the first and biggest sit at the table of TSMC they might still be thinking "Even with our own silicon design, we're still highly dependent on another company that's currently all but extorting more money out of everybody but us and might not even deliver on their performance promises"

I think that with all their incomprehensible quantity of money, their next logical step must be for Apple to start thinking if not their own fabs outright, someone they own and control exclusively. I know it seems implausible to impossible but again, it would take impossible amounts of investment and if anyone has that kind of money, it's Apple. In fact it's *only* Apple and almost nobody else.
If they were going to do that, they'd have already started by buying the smaller, cheaper suppliers, those for screens, speakers, PCBs, ece before buying a fab. Buyign a fab means being responsible for enw technologies in said fab, and I doubt apple wants to make that kind of investment.
 
If Apple’s generational improvements are going to be modest at best going forward, then its fanfare transition away from x86 might not look so good in hindsight. Especially given that Intel is promising some pretty solid performance increases with Alder Lake, with Raptor Lake to follow, and then Meteor Lake. And of course Zen4 and Zen5 ought to be pretty solid updates as well. It would be quite laughable if the incumbent x86 vendors deliver future solutions that are superior in every way to Apple Silicon… making everyone wonder why did Apple leave x86 again?

Of course, only time will tell and we have to wait for final silicon year after year to make any conclusive determinations. Will continue to monitor.
They still have the option to add more high performance cores and improve clock speeds even if the architecture won't improve that much. They are also usually the first to buy out new process nodes which gives them an advantage.
 
Even though Apple still has the first and biggest sit at the table of TSMC they might still be thinking "Even with our own silicon design, we're still highly dependent on another company that's currently all but extorting more money out of everybody but us and might not even deliver on their performance promises"

I think that with all their incomprehensible quantity of money, their next logical step must be for Apple to start thinking if not their own fabs outright, someone they own and control exclusively. I know it seems implausible to impossible but again, it would take impossible amounts of investment and if anyone has that kind of money, it's Apple. In fact it's *only* Apple and almost nobody else.

While their own Fab sounds like a good idea, history has shown that doing so can bring up issues. AMD's Fab started falling behind Intel in the 2000s, along with internal management issues and bulldozer being a doomed lineup for nearly an entire decade. AMD was tied to global foundries until they fell so far behind that they breached contract.

Intel is facing Fab problems.

Apple could dump Billions into new Fabs, and spend Billions more trying to keep up in Fab Tech. Just to run into a wall.

Honestly for Apple they are better off staying flexible and just allow TSMC, Samsung, and Intel to scale up and continue to use the best node available for ether vendor.


IMO Apple's biggest issue is they are dependent on their iPhone business. Loyalty only goes so far, and we all know how much can change in a decade. A massive drop in user base could easily spell doom for apple in terms of current business size. Unlike MS they really are not diverse.

Apple's VR goals may just end up like MS's
Apple's Car goals may never see light of day.


I'm more surprised Apple hasn't worked to be a bigger player in the cloud space. If there is one area Apple could easily focus money on for their own good. Cloud service. Apple is reliant of Amazon's AWS service for iCloud and most of their cloud services. Google clearly hasn't felt the need to really compete with Azure and AWS, there is room for another.
 
I think its pretty obvious the biggest cause is the losing talent point. Money can buy you a lot but if you cant keep talent its a moot point anyway. May apple be experiencing what IBM did?

Comparing against the competition is a bad idea. Intel tried that and it spectacularly backfired. What it tells consumers is "this really isnt an upgrade, so look at these other numbers!".
Mostly what happens is, people add Apple to their resume and then boom, now you're hot property and can get insane stock options at startups. Frankly, unless the project/product is groundbreaking, this is going to happen. Iphones have had nothing major since the X. Just that way. Unless of course you wanna settle down, rest and vest.
 
Back