Apple's A20 Pro Shatters Geekbench 7's Single-Core Record as the First 2nm iPhone Chip

Apple's A20 Pro is the first smartphone chip built on TSMC's 2-nanometer N2 process, and early benchmark submissions have handed it a title Apple could not claim a year ago: the highest single-thread score on Geekbench 7, the benchmark version Primate Labs released in July. The most widely cited submission put the chip at 4,006 single-thread and 11,460 multi-thread, according to Tom's Hardware. A later listing reached 4,042 single-thread, which Inven Global reported as the top single-core score on the new benchmark.

Related reading: TSMC to Lift 2nm and 3nm Output as AI Demand Outruns Supply · TSMC Posts a Sixth Straight Record Quarter as AI Demand Outruns Supply.

The chip powers the iPhone 18 Pro, iPhone 18 Pro Max, and the foldable iPhone Duo that Apple announced on September 9. The phones are no longer the story. What deserves scrutiny is the number, because it arrives attached to a benchmark that changed how it counts, hardware that has not reached buyers, and a comparison set that mixes phones with desktop parts.

A record on a benchmark that just changed

Geekbench 7 did not simply renumber itself. Primate Labs rebuilt the multi-core test so a workload only runs across multiple threads when the real application it models actually does, as the Geekbench blog explains. The browser test, for instance, no longer feeds the multi-core score because web browsers are mostly single-threaded. The suite added AV1 and Opus encoding, a Whisper speech-recognition workload, and a Game Physics test built on the Jolt engine. The calibration baseline moved from an Intel Core i7-12700 to an AMD Ryzen 7 7700.

The consequence, documented by AppleInsider and Macworld, is that Geekbench 7 and Geekbench 6 scores cannot share a chart. The same M4 iPad Pro that scored 3,719 single-core and 13,635 multi-core on Geekbench 6 returned 3,197 and 12,959 on Geekbench 7. The silicon did not slow down. The measuring stick changed.

When the A20 Pro's 4,006 appears beside an A19 Pro's 3,249, both are Geekbench 7 numbers, so the comparison is legitimate. Tom's Hardware's 23.3% single-thread and 27.1% multi-thread generational gains rest on Primate Labs' own earlier Geekbench 7 runs of the A19 Pro. What is not legitimate is carrying over the iPhone 17 Pro's familiar Geekbench 6 scores and calling the gap real. It is not the same test.

What the A20 Pro actually is

Apple's specifications, summarized by MacRumors and iPhone in Canada, describe a six-core CPU split into two “super cores” and four efficiency cores. Apple calls the super cores desktop-class and says they are up to 20% faster, with a peak clock of 4.93GHz, up from 4.26GHz on the A19 Pro. The GPU moves to seven cores and is up to 40% faster, with neural accelerators Apple says double 8-bit floating-point throughput.

The Neural Engine count doubles to a pair of 16-core units, 32 cores in total, which Apple says doubles on-device AI compute. Memory bandwidth rises 50% over the A19 Pro through what Apple calls the widest memory interface in an iPhone. The phones carry 12GB of RAM, visible in the Geekbench listings themselves.

One design choice matters more for sustained work than for a two-minute burst. Apple moved to a package, inspired by its M-series Mac chips, where the silicon die and the memory sit side by side rather than stacked (MacRumors). That lifts the memory out of the chip's thermal path and lets the die press directly against a vapor chamber with three times the surface area of the iPhone 17 Pro's. Apple says the combination produces up to 40% higher sustained performance than the iPhone 17 Pro, and twice that of the iPhone 16 Pro.

That 40% figure is where the coverage gets slippery. MacObserver argues the 40% number Apple attaches to the A20 Pro belongs to the GPU speed claim, while the vapor chamber's stated benefit is surface area and water cycling, not a performance percentage. Apple's own materials, as relayed by MacRumors and by TrendForce, do pair the redesigned chamber and packaging with a 40% sustained-performance gain. Two 40% claims sit in the same paragraph of the same announcement, which is likely why they are often merged into one. Read the sustained number as Apple's, not as an independently measured result.

Single-core is the headline. Multi-core is the fine print.

Single-core performance measures responsiveness: browsing, app launches, the bursty work that fills most phone use. Multi-core measures throughput, and there core count decides the outcome. Six cores cannot hide from sixteen.

On Geekbench 7, Tom's Hardware measured the A20 Pro's multi-thread score at 11,460. A 16-core Ryzen 9 9950X3D reached 30,428, an Intel Core i9-14900KS hit 21,145, Apple's own M5 scored 18,671, and the M4 managed 15,806. The gap is not close. A separate GIGAZINE analysis of Geekbench 6 data put the M5 Max's multi-core score near 29,000, though that figure comes from the older benchmark and should not be placed beside the Geekbench 7 numbers.

The A20 Pro does outpace some mainstream laptop silicon in multi-thread, edging the 10-core Intel Core Ultra 5 325, per Tom's Hardware. That is a respectable result for a phone. It is not the flex the single-core headline implies, and the distance between the two scores is the most useful thing a reader can take from the early data.

How the single-core lead compares

The single-thread advantage is the part of the story that holds up across sources. Geekbench 7 places the A20 Pro roughly 26% ahead of the Ryzen 9 9950X3D and 32% ahead of the Core i9-14900KS in single-thread, the two desktop flagships Tom's Hardware chose as yardsticks. Against its own family, the chip is about 7% faster than the M5 in single-thread and 43% faster than the M3 on the same benchmark.

Among phones, the field is further back than the marketing implies. Qualcomm's Snapdragon 8 Elite Gen 5 scored 3,047 and Xiaomi's XRing O3 scored 2,996 on Geekbench 7 single-thread, both landing near the level of Apple's own A18 Pro from two years ago. The margin over Huawei's Kirin 9050 Pro runs into the hundreds of percent. Inven Global separately reported a 4,042 single-core result and noted that it beat an M5 Max's 3,731, a comparison Digital Trends also made.

The plainest way to state it: Apple now ships a phone chip that beats high-end desktop processors in the one test that rewards a fast single core, while remaining a mid-tier laptop part once all cores are busy.

What the early results do and do not prove

Every number above comes from an unverified Geekbench listing uploaded shortly after launch. iClarified catalogued three Geekbench 7 runs of the iPhone 18 Pro and six of the Pro Max, with single-core results ranging from 3,927 to 4,015 on the smaller model and 4,017 to 4,042 on the larger one. The averages were 3,983 and 4,025. A single best run is not a chip's performance, and Apple's retail firmware may differ from the units that were tested.

The Geekbench 6.7 results add another data point and another conflict. GSMArena and Gizmochina pair the iPhone 18 Pro with 4,719 single-core and 12,677 multi-core, and the iPhone 18 Pro Max with 4,612 and 11,819. AppleInsider and Notebookcheck attribute the stronger 4,725 and 4,719 runs to the Pro Max instead. Device identifiers in early listings are easy to mix up, and this desk could not resolve which model owns which score. Gizmochina's AnTuTu listing for the Pro Max, an overall iOS V11 score of 3,551,073 against 2,275,118 for the iPhone 17 Pro Max, is a similar magnitude of jump.

The price correction

One widely repeated figure needs correcting. Pre-launch analysis from TrendForce and IDC, reported through 9to5Mac and others, floated an iPhone 18 Pro starting price near $1,249 and a Pro Max near $1,449. At launch, Apple set the iPhone 18 Pro at $1,199 and the iPhone 18 Pro Max at $1,299, each a $100 increase over the prior generation, according to Hot Hardware and AppleInsider. The predictions overshot. The memory-cost pressure behind them is real, and it is the same pressure likely to shape next year's pricing, but the launch numbers are lower than the leaks.

iOS 27 arrived on September 14, four days before the first Pro units reach stores on September 18. The foldable Duo, which also carries an A20 Pro, does not go on sale until October 23 at $1,999, so its sustained performance under a foldable chassis is still untested.

Apple now ships a phone chip that beats high-end desktop processors in the one test that rewards a fast single core.

Why 2nm matters beyond the scoreboard

The process node is the quieter half of the story. TSMC's N2 is the first generation to use gate-all-around nanosheet transistors, and the foundry has said the baseline technology can deliver 10% to 15% more speed at the same power, or cut power by roughly 25% to 30% at the same speed, compared with the N3E node behind the A19 Pro. Apple is the first company to put that node in a shipping phone, which matters for more than bragging rights. It secures early access to TSMC's best yields at a moment when memory prices and advanced packaging capacity are both tight. The side-by-side die-and-memory package follows the same logic. Apple is spending area and process cost to buy bandwidth and thermal headroom instead of chasing clock speed alone.

Whether the rest of the industry follows quickly is open. Qualcomm and MediaTek both have 2nm designs on their roadmaps, and a node transition this expensive tends to reach flagship phones first and everything else much later.

What to watch

The interesting question is no longer whether Apple leads in single-thread. It is whether the thermal redesign closes the gap between a peak score and the performance a phone can hold after ten minutes of gaming. Snapdragon and MediaTek will need a reply, and the answer will probably come from their own 2nm parts rather than from a re-run of these benchmarks.

For now, treat the A20 Pro as a genuine silicon milestone with a modest asterisk. The single-core record is real within its benchmark and its generation. The desktop-class framing is true only for the shortest bursts. The rest of the story will arrive when retail phones do, and when Geekbench 7 accumulates enough results for the numbers to mean something beyond a launch week.