A cluster of leaks landed across September 2026, and together they gave the clearest pre-launch look yet at the A20 Pro before Apple confirmed the chip. Four outlets, wccftech.com, Notebookcheck, shattered.io, and MacRumors, published overlapping but not identical claims about the A20 Pro, the chip that now powers the iPhone 18 Pro line and the iPhone Duo. The headline numbers: a 4.90GHz performance-core clock, a 7-core GPU that wccftech reported as a first for any iPhone chip, and a memory bus wide enough to beat every Apple silicon chip back through the M3. A September 11 benchmark leak reported by Gigazine then added the first real performance numbers to the mix: roughly 25% faster CPU performance and about 40% higher GPU performance than the A19 Pro, with the chip observed running at 4.93GHz. At the time, none of it came from Apple directly, all of it came from teardown photos, die shots, firmware strings, and a benchmark run that leakers had been picking apart since late August.
What makes this leak cycle different from the usual pre-launch noise is the level of disagreement between sources that are normally in lockstep. Wccftech pegged memory bandwidth at a precise 115.2GB/s. Notebookcheck and shattered.io both framed the gain as roughly 50% higher than the prior generation without committing to wccftech’s exact figure, a number close to what an earlier round of A20 Pro reporting also suggested for the process shrink itself. A separate MacRumors analysis, working backward from Apple’s already-shipped M6 chip, arrived at a more conservative estimate near 85GB/s. That spread, more than 30GB/s between the high and low estimates, is the story: multiple credible outlets are triangulating the same chip and landing in different places, which says as much about the leak ecosystem in 2026 as it does about the A20 Pro itself. Apple’s own September 2026 launch materials ultimately sided with the more conservative framing, confirming roughly 50% higher memory bandwidth than the A19 Pro without publishing an exact GB/s figure, validating Notebookcheck and shattered.io’s percentage-based estimate over wccftech’s precise 115.2GB/s claim, a figure PCMag’s own September 2026 review independently landed on as well, reporting a 50% increase in memory throughput for the A20 Pro.
Since this leak cycle played out, Apple has made the chip official. At its September 9, 2026 event the company unveiled the A20 Pro inside the iPhone 18 Pro and iPhone 18 Pro Max, announcing that it also powers the iPhone Duo, Apple’s first smartphone chip built on TSMC’s 2-nanometer process, per MacRumors’s coverage of the announcement, with preorders opening September 12 and retail availability on September 18. The reveal broadly confirmed the direction the leaks below had been triangulating: early post-launch analysis matched the most-corroborated pre-launch figures, including the shift to TSMC’s 2-nanometer process and the step up to a 7-core GPU, and Apple went further than any leak had, telling MacRumors the A20 Pro delivers 40% higher sustained performance than the A19 Pro and roughly 2x the sustained performance of the A18 Pro, official performance framing none of the pre-launch leaks had specifically claimed. The finer contested numbers, such as the exact memory bandwidth in GB/s, were still not spelled out in Apple’s own launch materials. The leak-by-leak breakdown that follows now reads as the pre-launch record this announcement settles.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the leaks actually claim about the A20 Pro
Strip away the outlet-by-outlet variation and a consistent core spec sheet emerges.
CPU: six cores, familiar layout, higher clocks
The A20 Pro keeps a 6-core CPU arranged as 2 performance cores plus 4 efficiency cores, the same split Apple used on the A19 Pro in 2025, a layout Tom’s Hardware’s September 2026 review of the shipping chip confirmed directly. Wccftech’s report put the performance cores at up to 4.90GHz and the efficiency cores at 2.60GHz, with 16MB of shared cache for the performance cluster and 8MB for the efficiency cluster.
Holding the core count steady while pushing peak clocks higher is a familiar Apple pattern: rather than adding cores every generation, Apple has generally leaned on higher clocks, larger caches, and process improvements instead. A bigger cache matters more as clocks climb, which is likely why the leaked 16MB/8MB split grew alongside the jump to 4.90GHz.
A September 11, 2026 benchmark leak reported by Gigazine added the first real-world performance data point on top of the spec leaks: the A20 Pro reportedly ran about 25% faster than the A19 Pro on CPU workloads, with the performance core observed hitting 4.93GHz during the benchmark run, a hair above wccftech’s earlier 4.90GHz spec-sheet figure, a gap consistent with the normal variance between a firmware-derived leak and a clock speed measured on an actual running chip. Retail-unit testing has since narrowed that figure: Tom’s Hardware’s September 2026 review clocked the A20 Pro’s performance cores at roughly 20% faster than the A19 Pro, below both the pre-launch leak and the Gigazine benchmark, a reminder that real-world CPU gains on shipping hardware tend to settle below the most optimistic early estimates.
GPU: seven cores, up from six
The GPU side is where the generational jump is bigger: a 7-core GPU, which wccftech reported would make the A20 Pro the first iPhone chip to carry a 7-core GPU, up from six on the A19 Pro, with each core reportedly clocked around 1.60GHz. A September 3, 2026 MacRumors report, citing an alleged solder-joint diagram, independently corroborated both the core count and the wider memory interface, making the 7-core GPU the most consistently corroborated figure in the entire leak set across four separate outlets, a figure Apple’s September 9, 2026 event then confirmed directly, and one PCMag’s own September 2026 coverage of the shipping A20 Pro verified again after launch.
On paper, the core-count jump alone is a one-core, roughly 17% increase. Shattered.io’s comparison table, detailed in the next section, went further and claimed a “40% faster graphics” performance figure that neither of the other outlets carried at the time. That gap between the core-count math and the performance claim looked hard to square until Gigazine’s September 11 benchmark leak put GPU performance at about 40% higher than the A19 Pro too, independent, benchmark-based support for a number that had spent over a week resting on a single outlet’s say-so, covered in more detail below. The gap would still only fully close if Apple also improved per-core efficiency or pushed clocks beyond the leaked ~1.60GHz figure, but the safest reading has shifted: both the GPU core increase and the roughly 40% performance gain now have independent corroboration from multiple leak sources, and Apple’s September 9 event has since confirmed the 7-core count directly, and Apple’s own launch materials went on to confirm the roughly 40% faster graphics performance figure as well, closing the loop on what had been the single biggest leak-to-official gap in the entire spec sheet.
Memory, packaging, and the process node
The memory subsystem is the most contested part of the leak. Wccftech, Notebookcheck, and now MacRumors all describe a 96-bit memory bus, a jump from the 64-bit bus Apple has used on iPhone chips for more than a decade. Notebookcheck’s motherboard-leak report tied that wider bus to 12GB of RAM and a packaging method called WMCM (wafer-level multi-chip module), which places the memory dies beside the SoC rather than stacked on top of it, a departure from Apple’s usual package-on-package approach. Shattered.io’s comparison table lists the same 96-bit bus alongside a claimed 32-core Neural Engine, a figure that had not been corroborated by wccftech, Notebookcheck, or MacRumors ahead of launch, but which Apple’s own September 2026 unveiling confirmed outright, with GadgetsNow describing the A20 Pro’s dual Neural Engine as totaling 32 cores and Apple saying it doubles Neural Engine compute versus the A19 Pro, which used a 16-core design.
On process node, the outlets agreed: the A20 Pro was expected to move from the 3-nanometer node used on the A19 Pro to TSMC’s 2-nanometer N2 process, a shift Apple confirmed at its September 9, 2026 event. Notebookcheck’s die-shot analysis estimated a die size around 98.6 square millimeters, a figure Apple has not commented on, consistent with the company’s decade-long practice of staying quiet on unreleased silicon’s finer specs until launch day and beyond.
The outlets, and where they disagree
Leak journalism runs on a mix of factory-floor photos, customs filings, and firmware strings pulled from developer betas, and no two outlets always source the same document. That is visible in how differently wccftech, Notebookcheck, shattered.io, and MacRumors each framed the same chip this month, and how a fifth outlet, Gigazine, later added an actual leaked benchmark run into the mix.
Wccftech: the most specific numbers
Wccftech was the most specific, publishing exact clock speeds (4.90GHz and 2.60GHz), an exact bandwidth figure (115.2GB/s), and a specific transfer rate (9,600MT/s) tied to LPDDR5X memory. The outlet’s framing was that 115.2GB/s would exceed the bandwidth ceiling of the M1, M2, and M3 chips, a notable claim given those are laptop and tablet-class silicon, not phone chips.
Notebookcheck: a more conservative bandwidth claim
Notebookcheck took a more conservative tone, describing the bandwidth gain as “up to 50%” rather than committing to a single number, and focused more of its reporting on the WMCM packaging change and the 12GB memory configuration.
That 50% figure is not far from wccftech’s implied gain: going from a 64-bit to a 96-bit bus is itself a 50% increase in bus width, so Notebookcheck’s framing may simply describe the same architectural change in relative terms rather than offer an independent bandwidth estimate. Read that way, Notebookcheck and wccftech agree more than the headline numbers suggest; the real daylight is between wccftech’s specific 115.2GB/s claim and MacRumors’ separate ~85GB/s estimate, covered below.
Shattered.io: the same core specs, plus two unconfirmed numbers
Shattered.io’s comparison table repeated the 7-core GPU and 96-bit bus claims but added the unconfirmed 32-core NPU figure and a “40% faster graphics” performance claim that neither of the other two outlets carried.
Both of shattered.io’s extra numbers were worth flagging precisely because they were extra: a figure only one outlet reports at all is a weaker signal than a figure two outlets report differently. The 32-core Neural Engine claim remains the least reliable figure in the leak set for that reason, a point revisited in the “what remains unconfirmed” section below. The 40% graphics claim has since gained independent support from Gigazine’s September 11, 2026 benchmark leak, covered later in this article, which moves it out of the single-source category even though Apple still has not confirmed it.
MacRumors: a solder-joint diagram, then a bandwidth estimate
MacRumors weighed in twice. On September 3, 2026, an alleged solder-joint diagram gave the outlet an independent read on the 7-core GPU and the wider 96-bit memory interface, corroborating wccftech, Notebookcheck, and shattered.io through a completely different sourcing method than the teardown and firmware-string approach the other three used. Two days later, MacRumors took a different approach entirely for bandwidth specifically, working backward from Apple’s M6 chip to estimate what a phone-class derivative might carry, landing on a bandwidth figure closer to 85GB/s, a full 30GB/s below wccftech’s number. That gap mattered for anyone trying to gauge what official specs would look like before Apple’s September 9 announcement, and while Apple’s own keynote confirmed a roughly 50% memory bandwidth gain over the A19 Pro, it still didn’t spell out an exact GB/s figure, leaving the wccftech-versus-MacRumors precise-number dispute technically unresolved. Readers should treat the CPU core count (6, split 2+4) and GPU core count (7) as the most consistently corroborated figures across outlets, and treat the exact bandwidth number in GB/s as the least settled.
A20 Pro vs A19 Pro: the generational gap
The comparison tables published by shattered.io and Notebookcheck both use the A19 Pro as the baseline, since it is the chip the A20 Pro succeeds. The table below consolidates the leaked figures against that known baseline.
| Spec | A19 Pro (2025, shipped) | A20 Pro (2026, leaked) | Source |
|---|---|---|---|
| Process node | 3nm (N3) | 2nm (TSMC N2) | shattered.io, wccftech |
| CPU core layout | 6-core (2P+4E) | 6-core (2P+4E), unchanged | wccftech, Notebookcheck |
| Performance core clock | Not disclosed by Apple | Up to 4.90GHz (leaked); 4.93GHz observed in benchmark | wccftech, Gigazine |
| Efficiency core clock | Not disclosed by Apple | Up to 2.60GHz (leaked) | wccftech |
| GPU cores | 6-core | 7-core (leaked) | wccftech, Notebookcheck, shattered.io, MacRumors |
| GPU clock | Not disclosed by Apple | ~1.60GHz per core (leaked) | wccftech |
| CPU performance gain (benchmark) | Baseline | ~25% faster (leaked) | Gigazine |
| GPU performance gain (benchmark) | Baseline | ~40% higher (leaked) | shattered.io, Gigazine |
| Memory bus width | 64-bit | 96-bit (leaked) | wccftech, Notebookcheck, shattered.io, MacRumors |
| Memory bandwidth | Not disclosed by Apple | 115.2GB/s (wccftech) vs ~85GB/s (MacRumors estimate) | wccftech, MacRumors |
| Memory configuration | 8GB (typical prior-gen) | 12GB (leaked) | Notebookcheck |
| Packaging | Package-on-package | WMCM, memory beside SoC (leaked) | Notebookcheck |
Reading the table: what’s corroborated vs. single-source
Not every row in that table carries the same weight. Six of the twelve specs, process node, CPU core layout, GPU core count, memory bus width, performance-core clock speed, and the GPU performance gain, are now reported consistently by two or more outlets, the closest thing to confirmation a leak gets before Apple’s own announcement. One row, memory bandwidth, has two outlets citing meaningfully different numbers. The remaining five rows rest on a single outlet each.
The single-source figures most likely to move
Single-source does not mean wrong, but it does mean these were the numbers to watch in the run-up to Apple’s iPhone 18 Pro unveiling on September 9, and the ones still worth scrutinizing now that some, like the exact bandwidth figure, remain unresolved even after launch. That list got shorter on September 11: Gigazine’s benchmark leak put the performance core at 4.93GHz, close enough to wccftech’s 4.90GHz figure to count as convergence rather than contradiction, giving that clock speed its first independent corroboration. The 2.60GHz efficiency-core figure still comes from wccftech alone, the 25% CPU performance gain comes from Gigazine alone, and the 12GB RAM figure and WMCM packaging detail both trace back to Notebookcheck specifically. Any additional outlet corroborating one of these would move it into the “reliable” column; contradicting one would put it in the same “disputed” category as the current bandwidth disagreement.
A20 Pro Benchmark Leak: 25% Faster CPU, ~40% Faster GPU
On September 11, 2026, Japanese outlet Gigazine reported the first leaked benchmark results for the A20 Pro, moving the leak cycle from spec sheets and teardown photos to actual measured performance. Gigazine’s source data put the A20 Pro at roughly 25% faster CPU performance and about 40% higher GPU performance than the A19 Pro, with the chip observed running at 4.93GHz, a hair above wccftech’s earlier 4.90GHz spec-sheet figure for the performance core. Separate A20 Pro prototype benchmarks that surfaced the same month scored 4,719 in Geekbench single-core testing and 12,677 in Geekbench multi-core testing, the first concrete third-party benchmark figures to accompany the percentage-based performance claims. Retail hardware has since posted slightly different numbers: Tom’s Hardware’s September 2026 review recorded a Geekbench 7 score of 4,006 single-core and 11,460 multi-core on a shipping iPhone 18 Pro, both below the earlier prototype figures and a reminder that pre-launch benchmark leaks don’t always match retail-unit performance exactly.
How the benchmark numbers line up with the spec leaks
The 4.93GHz reading is close enough to wccftech’s 4.90GHz figure that the two likely describe the same chip, with the small gap explained by the usual difference between a firmware-derived spec and a clock speed observed during an actual benchmark run. The roughly 40% GPU performance figure is more significant: it is the first independent, benchmark-based support for a number that had previously come only from shattered.io’s single-sourced “40% faster graphics” claim, discussed in the outlet breakdown above. The 25% CPU performance figure is new information entirely; none of the spec-leak outlets had previously estimated an overall CPU performance delta, only individual clock speeds.
What a single benchmark leak does and doesn’t prove
A single leaked benchmark run is a meaningfully different kind of evidence than a repeated spec claim, and it comes with its own caveats. Benchmark scores from pre-production silicon can shift before launch as Apple finalizes firmware and thermal tuning, and a single run says nothing about sustained performance under real-world thermal constraints inside a phone chassis. Weighed alongside the rest of the leak set, the benchmark strengthens the case that the underlying GPU and clock-speed leaks are in the right ballpark, without yet rising to the level of a confirmed spec.
How the leaked bandwidth stacks up against Apple’s own M-series chips
The single claim generating the most attention is wccftech’s assertion that A20 Pro bandwidth could exceed the M1, M2, and M3 chips, Apple’s own laptop and tablet silicon, not phone chips. That framing is worth unpacking, because Apple has historically kept a wide gap between iPhone and Mac memory subsystems specifically to manage cost, battery, and thermal envelope in a phone-sized enclosure. If the 115.2GB/s figure holds, it would represent one of the largest single-generation jumps in iPhone chip history, driven almost entirely by the wider 96-bit bus rather than a faster memory clock.
| Chip | Class | Memory bus | Approx. bandwidth |
|---|---|---|---|
| Apple M1 | Mac/iPad | 128-bit | Lower than leaked A20 Pro claim, per wccftech |
| Apple M2 | Mac/iPad | 128-bit | Lower than leaked A20 Pro claim, per wccftech |
| Apple M3 | Mac/iPad | 128-bit | Lower than leaked A20 Pro claim, per wccftech |
| Apple A19 Pro | iPhone | 64-bit | Baseline, not disclosed by Apple |
| Apple A20 Pro (leaked) | iPhone | 96-bit | 115.2GB/s (wccftech) or ~85GB/s (MacRumors estimate) |
What “beating the M3” would actually mean
Even at the more conservative 85GB/s estimate, a phone chip beating three generations of Mac-class silicon on raw memory bandwidth would be a notable milestone, and it lines up with a broader pattern: Apple has been steadily narrowing the architectural gap between its phone and laptop chips since the M-series debuted in 2020.
It is worth being precise about what this comparison claims. Wccftech’s assertion is about memory bandwidth specifically, not overall performance; the M1, M2, and M3 remain far ahead of any iPhone chip on sustained CPU and GPU workloads, since they run at higher power envelopes with active cooling a phone cannot match. A phone chip out-pacing a three-generation-old Mac chip on bandwidth is a real milestone, but a bandwidth milestone specifically.
Breaking down the bandwidth math: does 115.2GB/s actually check out?
The wccftech and MacRumors bandwidth figures cited throughout this leak cycle aren’t just competing claims to take on faith, both can be checked against the underlying bus-width and transfer-rate numbers each outlet reported alongside them.
The arithmetic behind wccftech’s number
Memory bandwidth isn’t a figure Apple invents out of thin air; it’s bus width multiplied by transfer rate, divided by eight to convert bits into bytes. Run wccftech’s own numbers through that formula and they hold up: a 96-bit bus works out to 12 bytes per transfer, and 12 bytes multiplied by an effective 9,600MT/s LPDDR5X transfer rate lands at exactly 115.2GB/s, 12 × 9,600,000,000 transfers per second equals 115,200,000,000 bytes per second, with no rounding required. That internal consistency doesn’t prove the underlying leak is correct, but it does rule out the more common failure mode in leaked spec sheets, where a bus-width figure and a bandwidth figure pulled from two different sources get stitched together and quietly don’t agree.
Running the same formula backward on MacRumors’ more conservative ~85GB/s estimate implies a transfer rate of roughly 7,083MT/s (85GB/s ÷ 12 bytes), noticeably below the 9,600MT/s figure wccftech cited for the same chip. That gap is likely the real source of the two outlets’ disagreement: it has less to do with whether the bus is 96-bit, which four separate outlets now agree on, and more to do with which memory-clock assumption each outlet built its estimate on.
Why the 50% figure shows up twice
It’s also worth noticing that Notebookcheck’s “up to 50%” bandwidth-gain estimate and the bus-width jump from 64-bit to 96-bit describe the exact same ratio: 96 divided by 64 is 1.5, or a 50% increase. If bandwidth scales linearly with bus width at an unchanged memory clock, a 50% wider bus predicts a 50% bandwidth gain on its own, with no faster memory chips required. Read that way, Notebookcheck’s figure may not be an independent bandwidth measurement so much as the same bus-width leak restated in different units, which would also mean the LPDDR5X transfer rate itself could be carrying over largely unchanged from the A19 Pro, with the entire generational bandwidth gain coming from the wider bus rather than faster memory.
For readers trying to decide which bandwidth figure to trust, the math points to a practical rule: weight the transfer-rate assumption more heavily than the bus width. The bus-width jump is the harder engineering claim to fake, since it shows up directly in die shots and solder-joint diagrams, which is why four outlets converge on 96-bit. The transfer rate is a firmware-tunable number that can still shift before Apple locks final specifications, and it’s exactly where the 115.2GB/s and ~85GB/s estimates part ways.
Why Apple would widen the memory bus now
On-device AI and memory supply chain pressure
None of the leaked reports specify Apple’s exact motivation, but the timing lines up with two publicly known pressures. First, Apple has spent much of 2026 building out on-device AI features under its Apple Intelligence branding, a push tied directly to the A20 Pro’s on-device AI performance, and larger, faster memory access directly benefits the kind of local model inference that runs image generation and language tasks on-device rather than in the cloud. Second, Apple has been diversifying its memory supply chain. The company locked in a multi-year NAND supply agreement earlier in 2026, and has separately faced reported DRAM shortages that pushed component costs higher across its lineup. A wider, more efficient memory bus reduces the amount of raw memory capacity needed to hit a given performance target, which can offset some of that cost pressure even as Apple adds RAM.
Memory bandwidth matters disproportionately for on-device AI because of how large language models actually run. Generating each word of output typically requires streaming the model’s full set of weights through memory at least once per token, which makes autoregressive text generation a memory-bandwidth-bound task rather than a compute-bound one on most modern chips. In practice, that means a 50% increase in memory bandwidth can translate more directly into faster on-device model responses than a similarly sized jump in raw CPU or GPU compute would, since the bottleneck for local inference is usually how fast data moves between memory and the processor, not how fast the cores can crunch it once the data arrives.
WMCM packaging: a thermal trade-off
The WMCM packaging change Notebookcheck reported also fits a broader industry shift. Placing memory dies beside rather than stacked on top of the main SoC is a technique that trades a small amount of board space for better thermal characteristics, since heat from the two dies does not compound vertically. That trade-off is more attractive in a phone, where thermal headroom is the tightest constraint on sustained performance, than it would be in a laptop or desktop chip.
The die shrink to 2-nanometer N2 likely helps absorb that trade-off: a denser process keeps the leaked 98.6 square millimeter die estimate in check even as the package spreads out sideways for the memory, one plausible reason both changes arrive in the same generation.
Competitive context: how A20 Pro’s leaked specs compare to rival flagship chips
CPU clocks: A20 Pro vs. Snapdragon 8 Elite Gen 6
The A20 Pro leak lands in a year when Apple’s mobile silicon lead has narrowed on paper, if not always in real-world benchmarks. Qualcomm has already confirmed its Snapdragon 8 Elite Gen 6 lineup, including dual-chip variants running up to 5GHz, ahead of Apple’s own announcement. At wccftech’s leaked 4.90GHz figure, since bolstered by Gigazine’s independent 4.93GHz benchmark reading, Apple is clocking slightly below Qualcomm’s fastest configuration, though Apple’s chips have historically extracted more performance per clock cycle thanks to a wider core design, so raw clock speed comparisons alone tend to understate Apple’s actual single-thread performance.
Measured strictly on peak clock, the gap is small: 4.90GHz against a dual-chip Snapdragon configuration running up to 5GHz is a difference of roughly 0.1GHz, or about 2%. That is close enough that architecture, not raw clock speed, will decide which chip actually feels faster day to day.
GPU cores: incremental gains vs. the core-count race
On the GPU side, a jump to 7 cores is incremental rather than dramatic, but it continues a pattern of Apple prioritizing sustained thermal performance over peak core counts. Rival Android flagship chips from MediaTek and Qualcomm have generally used higher core counts on paper while Apple has leaned on architectural efficiency. The wider memory bus, if confirmed, would be one of the more significant competitive moves in this generation, since memory bandwidth directly affects how well a phone chip handles the growing demands of on-device generative AI workloads that both Apple and its Android rivals are racing to support locally rather than through cloud APIs.
Historical accuracy of Apple chip leaks
Apple’s silicon leaks have a mixed track record heading into this cycle.
What tends to leak accurately
Core counts and process nodes leaked ahead of launch have generally proven accurate in recent years, largely because they can be sourced from TSMC supply chain filings and die shots that are hard to fake.
What tends to change before launch
Exact clock speeds and bandwidth figures have been less reliable, since those numbers often depend on final firmware tuning that Apple locks down closer to launch, and can vary between the exact chip Apple ships and pre-production samples that leak earlier in the cycle. That history is a useful lens for weighing the current leak set: the 6-core CPU layout, corroborated across multiple outlets, and the 7-core GPU count, now corroborated across four separate outlets using different sourcing methods (including a September 3, 2026 MacRumors report built on an alleged solder-joint diagram) and confirmed outright by Apple on September 9, sit in the “generally reliable” category. The September 11, 2026 Gigazine benchmark leak adds a data point in between: its 4.93GHz reading lines up closely with wccftech’s 4.90GHz clock figure, and its roughly 40% GPU performance gain matches shattered.io’s earlier single-sourced claim, but benchmark runs on pre-production silicon can still shift before Apple locks final firmware. The exact 115.2GB/s bandwidth figure, sourced to a single outlet and already contradicted by a second outlet’s independent estimate, sits in the “treat with caution” category until Apple confirms it or a fourth independent source corroborates it.
What this means for iPhone 18 Pro buyers
For anyone who picked up the iPhone 18 Pro since it went on sale September 18, or is still deciding, the practical takeaway from this leak cycle is narrower than the headline numbers suggest.
Gamers and on-device video work
The GPU core increase from 6 to 7, confirmed at Apple’s September 9 event, translates into a modest but real improvement in sustained graphics performance, particularly useful for gaming and on-device video work.
On-device AI users
The memory bus widening matters most for on-device AI features, where more bandwidth means the phone can run larger local models, or run existing models faster, without offloading as much processing to the cloud.
Camera and display buyers
Buyers upgrading primarily for camera or display changes should treat the chip leak as a secondary factor, since Apple’s silicon gains in recent generations have shown up more in efficiency and sustained performance than in headline peak-speed numbers that rarely matter in everyday use. None of the leaked figures in this article change that calculus: the chip changes are real, but for now they show up as background improvements to efficiency and sustained performance rather than as something a buyer would notice in a store demo.
Where the A20 Pro Ships Beyond the iPhone 18 Pro
Apple announced in September 2026 that the A20 Pro powers the iPhone 18 Pro, the iPhone 18 Pro Max, and the iPhone Duo. Apple then announced pre-orders for the foldable iPhone Duo and related accessories on October 16, 2026, and the iPhone Duo reached Apple Store locations worldwide on October 23, 2026. For buyers weighing the leaked figures, that means the chip is not limited to the two Pro phones, and the foldable is another way to get it.
Apple’s developer page says iPhone Duo will be available to customers starting October 23, 2026, and tells developers to recompile existing apps with Xcode 27.1, which adds development support for iPhone Duo, and to use Device Hub to visualize how apps behave on iPhone Duo across all poses and orientations. Apple’s own feature requirements also list the iPhone Duo next to the iPhone 18 Pro models: expressive customizable Siri voices require iPhone Duo, iPhone 18 Pro models, iPhone Air, iPhone 17 Pro models, or iPad models with the M4 chip or later with a minimum of 12GB of unified memory, and are available in English only. On the supply side, Apple is committing up to $100 million over the next three years with Perryman, which will supply titanium for products such as iPhone Duo.
The comparison in this article’s title still needs care. No available source supplies a dated, standardized A20 Pro-versus-M3 benchmark, and the results compare the A20 Pro with the A19 Pro, not with the M3. Readers should treat the beats-M3 framing as a memory bandwidth claim that came from leaks, not as an established conclusion about overall performance.
Market and supply chain implications
TSMC’s 2-nanometer ramp
The move to TSMC’s 2-nanometer N2 process, confirmed at Apple’s September 9 event, makes the A20 Pro among the first mass-market smartphone chips built on that node, part of a broader wave of AI chip development reshaping the mobile silicon market this year, ahead of most Android rivals. TSMC has been ramping N2 capacity through 2026 as part of its broader expansion, and Apple has historically been the anchor customer that absorbs a large share of a new node’s earliest, most expensive capacity, a position that tends to squeeze pricing and availability for other chipmakers competing for the same fab slots in the following two to three quarters. Now that the A20 Pro has shipped, expect continued reporting on N2 capacity constraints affecting other fabless chip designers through late 2026 and into 2027 if iPhone 18 Pro demand stays strong.
The cost of 12GB RAM across the lineup
The reported shift to 12GB of RAM, up from typical prior-generation configurations, also has supply chain implications. Apple has faced reported memory cost pressure in 2026, and a jump in per-unit RAM capacity across its entire Pro iPhone lineup would meaningfully increase Apple’s exposure to DRAM pricing, particularly if memory prices remain elevated into the device’s on-sale window.
That RAM increase compounds with the other changes described above: a wider 96-bit bus, a new WMCM packaging step, and a jump to 12GB of memory are three separate cost pressures landing on the same chip in the same generation. None of the leaks so far address how Apple might offset that combined cost.
Predictions: what happens next
- The CPU core layout (6-core, 2P+4E) stayed consistent across every outlet through the rest of the leak cycle, and the 7-core GPU figure, corroborated by four independent outlets before launch using different sourcing methods, turned out to be one of the more durable numbers, with Apple’s September 9 event confirming the GPU core count directly.
- Expect continued disagreement on the exact bandwidth figure, since Apple’s own keynote and launch materials didn’t spell out a GB/s number, the real figure likely still lands somewhere between MacRumors’ ~85GB/s estimate and wccftech’s 115.2GB/s claim until a teardown lab or third-party test settles it.
- The wider memory bus lines up most naturally with on-device AI messaging, faster local model inference rather than a gaming or benchmark upgrade, which fits the on-device AI push Apple has leaned on across its 2026 marketing.
- Expect rival chipmakers, particularly Qualcomm, to continue publishing clock-speed comparisons now that Apple’s own numbers are out, a pattern already visible with Qualcomm’s Snapdragon 8 Elite Gen 6 disclosures earlier this year.
- Gigazine’s September 11, 2026 benchmark leak delivered on the pattern seen in prior iPhone chip cycles, where a benchmark score from a pre-production unit surfaces a few weeks after the initial spec leaks; expect additional independent benchmark results now that retail iPhone 18 Pro units are in reviewers’ hands following the September 18 on-sale date.
What remains unconfirmed
Several figures circulating around the A20 Pro leak deserve more skepticism than others.
The 32-core Neural Engine claim
Shattered.io’s claimed 32-core Neural Engine went uncorroborated by wccftech, Notebookcheck, or MacRumors through most of the leak cycle, and NPU core counts in particular have historically been among the least reliable pre-launch leaks, since Apple rarely discloses that architecture until launch day. That changed at Apple’s September 2026 unveiling, where the company confirmed a 32-core Neural Engine, double the 16-core design used in the A19 Pro, turning what had been the leak set’s shakiest number into one of its most decisively settled.
The 9,600MT/s transfer rate
The exact memory transfer rate of 9,600MT/s cited by wccftech is also a single-source figure. Combined with the 115.2GB/s bandwidth claim built on top of it, this is the single number in the entire leak set most likely to still be revised, either up or down, since Apple’s own September 9 keynote didn’t spell it out. Readers should treat anything not corroborated across at least two independently sourced reports as provisional until Apple publishes final specifications or a teardown lab measures the number directly.
Frequently asked questions
Has Apple confirmed any A20 Pro specifications?
Yes, partially. Apple confirmed the A20 Pro at its September 9, 2026 event, unveiling it inside the iPhone 18 Pro and iPhone 18 Pro Max, and early post-launch analysis matched the most-corroborated pre-launch figures, including the shift to TSMC’s 2-nanometer process and the step up to a 7-core GPU. Apple’s own launch materials didn’t spell out finer numbers like exact clock speeds or memory bandwidth in GB/s, so figures like the 4.90GHz clock speed and 115.2GB/s bandwidth claim remain sourced to leaks, wccftech specifically, rather than to Apple directly.
What is the leaked A20 Pro CPU clock speed?
Wccftech reported performance cores running up to 4.90GHz and efficiency cores up to 2.60GHz, across a 6-core layout split 2 performance and 4 efficiency cores, unchanged from the A19 Pro’s core count. A September 11, 2026 Gigazine benchmark leak separately observed the performance core running at 4.93GHz, closely matching wccftech’s figure.
How many GPU cores does the A20 Pro have?
Four separate outlets (wccftech, Notebookcheck, shattered.io, and MacRumors) independently reported a 7-core GPU, which wccftech said would be a first for any iPhone chip, up from 6 cores on the A19 Pro. MacRumors’ September 3, 2026 report cited an alleged solder-joint diagram, a different sourcing method than the other three outlets used, and Apple confirmed the 7-core GPU directly at its September 9, 2026 event.
How much faster is the A20 Pro than the A19 Pro?
A September 11, 2026 Gigazine benchmark leak put the A20 Pro about 25% faster than the A19 Pro on CPU performance and about 40% higher on GPU performance. Those figures come from a single leaked benchmark run on pre-production silicon and have not been confirmed by Apple, though the GPU figure lines up with an earlier, separately sourced claim from shattered.io.
Is the A20 Pro’s memory bandwidth really faster than the M3?
That is wccftech’s specific claim, citing a 115.2GB/s figure tied to a new 96-bit memory bus. A separate MacRumors estimate, working backward from Apple’s M6 chip, put A20 Pro bandwidth closer to 85GB/s, so the exact figure remains unsettled between outlets.
Does the leaked 115.2GB/s bandwidth figure actually add up?
Yes, mathematically. A 96-bit memory bus works out to 12 bytes per transfer, and 12 bytes multiplied by wccftech’s cited 9,600MT/s LPDDR5X transfer rate equals exactly 115.2GB/s, with no rounding needed. That confirms the figure is internally consistent, though it doesn’t independently verify that 9,600MT/s is the real transfer rate Apple shipped, which is where MacRumors’ more conservative ~85GB/s estimate, implying roughly 7,083MT/s, parts ways with wccftech’s version.
What manufacturing process does the A20 Pro use?
The A20 Pro moved to TSMC’s 2-nanometer N2 process, up from the 3-nanometer node used on the A19 Pro. Wccftech and shattered.io reported the shift ahead of launch, and Apple’s September 9, 2026 event confirmed it.
Which phone does the A20 Pro ship in?
The A20 Pro ships inside the iPhone 18 Pro and iPhone 18 Pro Max, confirmed at Apple’s September 9, 2026 event and on sale since September 18. Apple announced in September 2026 that the chip also powers the iPhone Duo, which reached Apple Store locations worldwide on October 23, 2026. Notebookcheck’s earlier die-shot report also referenced Apple’s first foldable iPhone as a potential device using related silicon, though that association was less firmly established than the iPhone 18 Pro connection at the time.
How much RAM is the A20 Pro expected to carry?
Notebookcheck’s motherboard-leak report cited 12GB of RAM, paired with the new WMCM packaging method that places memory dies beside the SoC.
When did Apple officially confirm A20 Pro specifications?
Apple confirmed the A20 Pro on September 9, 2026, alongside the iPhone 18 Pro and iPhone 18 Pro Max unveiling, consistent with prior iPhone chip cycles where official confirmation arrives with the iPhone hardware itself rather than in a standalone chip announcement.
