A20 Pro Geekbench 6: 4,200-4,400 single, 10,500-11,500 multi — 2nm TSMC, 42-48 TOPS

Apple will unveil the A20 Pro chip on Wednesday, September 9, 2026, and benchmarks will follow within days of the keynote. The A20 Pro is the first Apple-designed silicon built on TSMC's N2 (2nm) process and the first to use TSMC's Wafer-Level Multi-Chip Module (WMCM) packaging. This article lays out what the A20 Pro will most likely deliver in raw benchmarks, based on the A19 Pro as the baseline and the publicly documented characteristics of TSMC N2.

This is a projection, not a leak. The A20 Pro benchmark numbers will not be confirmed until after the September 9 keynote and the first wave of independent reviews in late September. The framework below is the best read on what to expect, with confidence ranges based on the available evidence.

A19 Pro baseline: where we are today

The A19 Pro, which shipped in the iPhone 17 Pro and iPhone 17 Pro Max in September 2025, is the most recent Apple Pro chip and the direct baseline for A20 Pro. Public Geekbench 6 submissions for the A19 Pro show:

• Single-core: ~3,800-3,900
• Multi-core: ~9,500-10,000

These are meaningful gains over the A18 Pro in the iPhone 16 Pro (single-core ~3,400, multi-core ~8,400), and consistent with Apple's typical generational improvement of 10-15% single-core and 12-18% multi-core on the Pro chip line. The A19 Pro is built on TSMC's N3P process (3nm-class, second-generation 3nm), not on the cutting-edge node — Apple held back N2 for A20 Pro specifically.

TSMC N2: what changes versus N3P

TSMC's N2 is the first mass-production node to use Gate-All-Around (GAA) nanosheet transistors, replacing the FinFET architecture that N3 and earlier nodes used. Three publicly documented differences versus N3P:

1. Logic density: ~20-25% higher at iso-performance. More transistors per square millimeter, which Apple can use for either more cores, more cache, or a smaller die.
2. Performance at iso-power: ~10-15% higher at the same clock speed, or the same performance at lower power. Apple typically takes the second path — same performance at lower power, which translates to longer battery life.
3. Power at iso-performance: ~25-30% lower. This is the headline efficiency number. For a chip running the same workload as A19 Pro, A20 Pro on N2 should consume 25-30% less power, all else equal.

The actual Apple advantage depends on how Apple uses the extra density (more cache, more cores, or both) and how aggressive the company is with clock speeds. Historical Apple nodes have not hit the maximum theoretical gain — Apple typically captures 50-70% of the theoretical node improvement in shipped silicon.

What to expect: A20 Pro projected benchmarks

Based on the A19 Pro baseline, the N2 node characteristics, and Apple's historical A-series generational gains, the realistic projection ranges are:

Benchmark	A19 Pro (actual)	A20 Pro (projected range)	Confidence
Geekbench 6 single-core	~3,800-3,900	4,200-4,400	High (10-15% gain is the historical pattern)
Geekbench 6 multi-core	~9,500-10,000	10,500-11,500	Medium (depends on core count vs clock)
GPU compute (Geekbench Metal)	~45,000	52,000-58,000	Medium (A-series GPU typically gains 15-25%)
Neural Engine TOPS	35 TOPS	42-48 TOPS	Medium (Apple doubles roughly every 2 generations)
Power efficiency (same workload)	baseline	+20-25%	High (N2 theoretical is 25-30%)

The single-core projection is the most reliable: Apple has consistently hit 10-15% single-core gains on the Pro chip line for six generations. The multi-core projection depends on whether Apple adds cores (unlikely on a phone SoC) or relies on clock speed improvements (more likely). GPU and Neural Engine gains vary more generation-to-generation.

What changes with WMCM packaging

The A20 Pro is also the first Apple chip using TSMC's Wafer-Level Multi-Chip Module packaging. WMCM allows the A20 Pro logic die and the DRAM to be packaged in a single integrated module, rather than as separate dies on a substrate.

For benchmarks, the practical implication is:

• Lower memory latency. WMCM allows shorter interconnect paths between logic and DRAM. This typically shows up as 5-10% improvement on memory-bound workloads (large language model inference, photo editing, video encoding).
• Better sustained performance. Discrete SoCs throttle under sustained load as the chip heats up. WMCM's tighter integration can run at higher performance for longer, which will show up in sustained Geekbench Metal scores rather than peak scores.
• Same total DRAM. The A20 Pro has 12GB of LPDDR5X DRAM — the same as A19 Pro in the Pro lineup. WMCM is about how the DRAM is connected, not how much there is.

How to read Apple's keynote benchmarks

Apple typically opens its September iPhone event with a chip performance slide. Three things to look for in the slide:

1. Single-core improvement over A19 Pro. Apple usually quotes this directly. If they say "20% faster single-core" that would beat the historical average and indicate Apple took more aggressive advantage of N2. If they say "10-15%," that is the standard pattern.
2. Neural Engine TOPS. Apple's on-device AI pitch has relied on Neural Engine TOPS since iPhone 15 Pro. The A19 Pro is 35 TOPS; the A20 Pro should be in the 42-48 TOPS range. Apple will likely quote a 30-40% improvement.
3. Power efficiency claim. Apple almost always quotes "X% less power for the same performance" versus the prior generation. For N2, expect 20-25% — Apple's usual conservative framing of TSMC's theoretical 25-30% node gain.

When to look for independent benchmarks

Apple's keynote is Wednesday, September 9. Independent benchmarks will follow in waves:

• September 9-10: First leaks and pre-release unit benchmarks from Twitter leakers and supply-chain contacts.
• September 18 onwards: First retail units ship. Geekbench submissions from reviewers and early buyers appear in the public database within 24-48 hours.
• Late September: MacRumors and 9to5Mac published reviews include standardized benchmark suites.
• October: AnandTech-style deep dives with sustained-load and per-component breakdowns.

For the most reliable benchmark numbers, look for the late-September / early-October reviews, not the immediate post-launch leaks. The leaks give you a number; the reviews give you context.

What about the A20 (non-Pro)?

Apple typically uses a slightly cut-down A20 (non-Pro) chip in the standard iPhone 18 and iPhone 18e. The non-Pro A20 will share the same N2 process and WMCM packaging as the A20 Pro but with:

• One fewer GPU core (5-core GPU vs 6-core on Pro)
• Slightly lower clock speeds on the CPU
• The same Neural Engine (Apple does not usually cut the Neural Engine on the non-Pro chip)

For most users, the A20 and A20 Pro will feel similar in daily use. The Pro chip pulls ahead in sustained GPU workloads (gaming, video editing, ML inference) and on the benchmarks.

Bottom line

The A20 Pro is the most consequential Apple chip since the M1 in 2020 — it is the first Apple silicon on a new transistor architecture (GAA nanosheet), and the first to use WMCM packaging. The realistic projection is 10-15% single-core improvement, 15-20% multi-core improvement, and 20-25% efficiency gain over the A19 Pro. Apple has used N2 specifically to land this generation, and the September 9 keynote should validate the projections above within the first 30 minutes of the event.

If you are deciding between iPhone 18 Pro and iPhone 17 Pro based on performance, the realistic answer is that the A20 Pro is meaningfully faster but not dramatically so. The Pro Max's variable-aperture camera, the 12GB RAM across the lineup, and the iOS 27 software features will likely matter more for day-to-day experience than the chip-level improvement alone.