Apple, in a first, reported no CPU gains for the A15, likely because of the engineer exodus to companies like Nuvia and Rivos
Apple has been long hailed for having the best CPU cores for consumer workloads for years. They have by far the highest performance per clock and efficiency driven …
The break is that Apple itself reported no CPU gains for the A15 — a first — with SemiAnalysis attributing it to engineers leaving for startups like Nuvia and Rivos. Independent testing complicated the picture further: reviewers found [[a:971437|efficiency gains and speed improvements ranging from 2.5% to 37% over A14 depending on workload]], while GPU performance actually jumped 52% against a five-year average of 19%.
First-order effects
Apple's positioning shifts mid-generation: with CPU leadership no longer the headline, the A15 pitch leans on the 52% GPU jump, the Neural Engine, and the efficiency wins the independent review measured.
Nuvia and Rivos get de facto validation — a flat Apple CPU generation is the strongest possible evidence that their poached engineering teams took real capability with them.
Second-order effects
Competitors building custom cores gain a credible window: the A14 analysis framed Apple Silicon as closing on x86 incumbents, but a stalled generation narrows that gap and gives Qualcomm-class players room to reposition around ex-Apple talent.
Benchmark scrutiny intensifies across the industry — once Apple concedes a flat year, buyers and reviewers will hold every vendor's generational claims to the same workload-by-workload accounting rather than headline percentages.
Third-order effects
If engineer mobility toward well-funded silicon startups continues, chip differentiation becomes a talent-retention contest rather than a process-node race, weakening any single vendor's durable core-design lead.
The pattern points toward a fragmented custom-silicon landscape: more companies capable of competitive cores means Apple's historical performance-per-clock moat becomes one strong design among several.
The trend: Custom CPU design is shifting from a moat held by one company to a contested field where engineer migrations between Apple-class teams and funded startups set the pace of generational gains.
Unprecedented in the whole history of iPhone, it wouldn't be concerning usually since they didn't change the process node... but that did happen in the past and still had CPU gains even on the same node. Apple chip supremacy potentially ending, ironic now that they are on Mac too…
The new A15 Bionic processor has 15 billion transistors and gets new AI and graphics abilities for better performance shooting photos and running apps, Apple says at #AppleEvent. https://www.cnet.com/...
“We believe #Apple had to delay the next generation CPU core due to all the personnel turnover Apple has been experiencing” https://semianalysis.substack.com/ ...
@dylan522p This was less about struggling and more about being second gen 5nm, essentially no process gains. The pattern when jumping to a new process is generally 20% perf gains so wouldn't expect much if staying on same process as they are.
“Frankly, the competition is still playing catch up to our chips. Not just from last year, but from two years ago” Says Apple as they introduce the new A15 Bionic chip #AppleEvent https://twitter.com/...
All the little details about the #A15Bionic from the presentation. Apple's comparing it vs the competition and not the iPhone 12, which is interesting. Wonder what the year vs year upgrades are. #AppleEvent https://twitter.com/...
@BenBajarin Process node isn't what dictates CPU architecture. It helps a bit on power and clocks, sure but it isn't the main reason. Apple's lead comes from architecture. They have the transistor budget to change the architecture and get more gains. They didn't use it on the CPU…
Low performance gain is likely due to the overheating issues that force Apple to throttle down the performance. I mentioned this in my previous leaks. If Apple couldn't bring vapor chamber cooling to the iPhone 13, performance would be limited again. We saw that with the (1/2) ht…