MacBook Neo Deep Dive: Benchmarks, Wafer Economics, and the 8GB Gamble

Updated: May 8th, 2026 with pricing and availability update. 

Preface: I’m not really a Mac guy. But I have deep respect for what Apple has done with their silicon, and I’ve been following their CPU journey since the Motorola 68k days through PowerPC, the Intel transition, and now their in-house Apple Silicon. What they’ve accomplished in the last five years is genuinely remarkable. Apple is one of the few original tech companies that has survived and thrived over the decades while still staying in the consumer tech space.

As a kid I used both Apple and Compaq computers in the text based OS days. Over the years I’ve purchased Apple systems periodically over the years and their recent entrants are extremely capable. While in the modern era I’ve never fully made the switch away from Windoze and Linux, I do give Apple props for doing what they do. 💪

The following (attempted) analysis hits close to home for me. AnandTech was one of my go-to sites when I built my first PC back in the day: a Tyan motherboard, Pentium II 233 MHz, SCSI hard drive, with Anand’s articles as the guide (I was a sophomore in HS I believe, and Anand was young as well). That was a blast, and Anand’s deep-dive hardware coverage was a huge part of what made the hobby so rewarding. (Anand eventually joined Apple, which tells you something about the caliber of talent they attract.) AnandTech had some of the best Apple silicon analysis ever published, and this article is written in that spirit: real data, real math, no fluff. 

With all that said, here’s a look at how Apple, and really only Apple, can deliver this kind of vertically and horizontally integrated product at a $599 price point while maintaining a comparatively high-quality build. They designed the chip, they control the OS, they negotiate directly with TSMC, and they amortize silicon costs across 230 million iPhones a year. Nobody else has that supply chain.

Yes, 8GB of RAM is a real limitation. But give it a year and the next version will almost certainly ship with 12GB and a modest CPU bump. Apple will maintain their margins, the world will continue on, and early adopters will have gotten a surprisingly capable machine in the meantime. There’s also a silver lining to the tight memory envelope: Apple has to keep macOS running well within 8GB, which is actually a nice forcing function against bloat and inefficiency. We could all use a little more of that.

Technical Analysis

---

On March 4, 2026, Apple unveiled the MacBook Neo, its most affordable Mac laptop ever at $599. The headline spec that has the internet arguing: instead of an M-series chip, the Neo runs the A18 Pro, the same processor from the iPhone 16 Pro.

“An iPhone chip in a Mac” sounds like a downgrade. The benchmarks tell a very different story.

The A18 Pro’s single-core performance lands between the M3 and M4, demolishes Intel and Qualcomm competitors at this price tier by 38-43%, and does it all in a fanless aluminum chassis with 16 hours of claimed battery life. The chip is not the constraint. The 8GB of RAM, with no upgrade path, is.

This article covers everything: actual benchmark data, how the A18 Pro compares to M-series chips architecturally, the wafer economics that make $599 possible, and why the global RAM shortage makes Apple’s timing look less like luck and more like strategy.

MacBook Neo Specs

The MacBook Neo is a 13-inch aluminum notebook built around the A18 Pro, fabricated on TSMC’s second-generation 3nm process (N3E). Here are the specs that matter:

Component	Specification
CPU	6-core: 2 performance (4.04 GHz) + 4 efficiency (2.42 GHz)
GPU	5-core Apple GPU, hardware ray tracing
Neural Engine	16-core, 35 TOPS (Apple Intelligence supported)
Memory	8GB unified LPDDR5x (soldered, no upgrade)
Storage	256GB ($599) or 512GB + Touch ID ($699)
Display	13″ Liquid Retina, 2408×1506, 500 nits
Ports	1x USB-C 3 (10 Gbps) + 1x USB-C 2 (480 Mbps) + 3.5mm
Battery	36.5Wh, up to 16 hrs video / 11 hrs web
Weight	2.7 lbs, fanless
Colors	Silver, Indigo, Blush, Citrus

To hit $599, Apple cut: MagSafe, Thunderbolt, backlit keyboard, haptic trackpad, P3 wide color, True Tone, Wi-Fi 7, and the 12MP webcam (replaced with 1080p). Touch ID is only on the $699 model. One of the two USB-C ports runs at USB 2.0 speeds, which is not good in 2026.

---

Hands-On: Three Thermal States

Every benchmark number you see in reviews is a snapshot of one moment. One ambient temperature, one background load, one thermal state. That is not how laptops work in real life.

So I ran Geekbench 6 on my own MacBook Neo under three different conditions, measuring what actually happens when you push a fanless 6-core chip past its comfort zone. The results were dramatic.

Test Setup

The machine: MacBook Neo (Mac17,5), Apple A18 Pro, 8GB unified memory, 256GB SSD, macOS Tahoe 26.3.2. All tests run on the same unit within the same 12-hour window.

Three conditions, tested in this order:

1. Cold start (fan-assisted): Machine rested overnight, then placed on a USB desk fan to keep the chassis at ambient temperature. Claude Code and screen sharing disabled. Three consecutive runs with 2-minute cooldowns between each.
2. Dev workload (Claude Code active): Cold start, but with Claude Code (Opus 4.6, 1M context) running in the background. This represents a real developer workflow: an AI coding assistant consuming memory and occasional CPU while you try to get work done.
3. Post thermal soak: After a 5-minute all-core stress test that drove CPU utilization to 570% and triggered aggressive thermal throttling. This is the worst case: what your Neo delivers after sustained heavy lifting.

Geekbench 6 Across Three States

Condition	Single-Core	Multi-Core	SC vs Cold Start
Thermal soak (5 min all-core stress)	476	1,340	-87%
Dev workload (Claude Code active)	709	1,305	-80%
Cold start (fan-assisted, 3-run avg)	3,569	8,879	Baseline

Read those numbers again. The same chip that posts 3,569 single-core when cold delivers 476 after five minutes of sustained load. That is an 87% reduction in single-core performance on the same hardware, running the same benchmark, separated by nothing but heat.

The cold start numbers (3-run average: SC 3,569, MC 8,879) match published A18 Pro scores almost exactly. The variance across three pristine cold runs was just 7 points on single-core, confirming the test methodology is sound. (Run 1, Run 2, Run 3)

One detail worth noting: multi-core scores under dev workload (1,305) and thermal soak (1,340) are essentially identical. Once the Neo hits its thermal or memory ceiling, multi-core performance converges regardless of the cause. The chip has one sustained performance floor, and both conditions find it.

The 60-Second Thermal Cliff

To understand why the post-soak score is so low, I ran a 5-minute all-core stress test and logged CPU utilization every 15 seconds.

For the first 60 seconds, the A18 Pro runs at full tilt: all six cores near 100%, CPU utilization around 570%. Then the thermal wall hits. Between T+60 and T+75, utilization crashes from 570% to 207%, a 64% drop in 15 seconds. For the remaining four minutes, the chip bounces between 188% and 360%, never recovering its burst performance.

There is one interesting spike at T+240 (448%) where the SoC briefly attempts to boost before throttling right back down. The cooling system simply cannot dissipate the heat fast enough to sustain high clocks.

This matches what Technetbook found independently: the A18 Pro hits its 105°C thermal limit and drops from 3.3 GHz to approximately 2.3 GHz. Modders have confirmed the cooling is the constraint: TweakTown measured an 18% Geekbench improvement with liquid cooling, and Hackaday documented doubled gaming frame rates with a water cooling mod.

Here is what the outside of the machine tells you during all this: I measured 97.6°F (36.4°C) on the hottest spot of the case surface with an infrared thermometer during sustained load. That is barely above body temperature. The chip is internally at 105°C and shedding 87% of its performance while the chassis feels perfectly comfortable in your lap. Apple made a deliberate design choice: comfort over sustained power.

What This Means in Real Use

The MacBook Neo is a sprinter, not a marathon runner. For tasks that complete within 60 seconds (compiling a small project, processing a batch of photos, rendering a short video clip), you get desktop-class single-core performance that beats Ryzen 9 chips. For tasks that sustain heavy load beyond a minute (long video encodes, large builds, training loops), you get dramatically less.

This is not a flaw. It is a design choice inherent to every fanless laptop, and the Neo makes that tradeoff at $599. The question is whether your workload fits inside the burst window. For the vast majority of users (web browsing, office work, light development, media consumption), every interaction is a burst: a page load, a document save, an app launch. Those users will never see the thermal wall.

For the full benchmark comparison including third-party data from every major competitor, see the benchmark tables below. For the hands-on review, including why this machine reminds me of a legendary computer from 25 years ago, see our MacBook Neo Review.

---

CPU Benchmarks: The Data

Actual Geekbench 6 results for the MacBook Neo were published by MacRumors on March 5, 2026. The Neo scored 3,461 single-core, 8,668 multi-core, and 31,286 Metal (GPU).

Here’s how that stacks up against both Apple’s own lineup and the $600-class Windows/ARM competition:

The single-core story is remarkable. The A18 Pro at 3,461 is 47% faster than the M1 (2,346), outperforms the M2 and M3, and lands within 6-7% of the M4 (3,696). Against the competition available at $600, it beats Intel’s Lunar Lake Ultra 5 226V by 38% and the Snapdragon X Plus by 43%. Only the unreleased Snapdragon X2 Plus (3,311) gets close, and it’s not shipping in sub-$700 laptops yet.

For the tasks this machine is built for (web browsing, documents, streaming, light photo editing), individual core performance a huge factor. With two performance cores the Neo will feels surprisingly snappy and the remaining efficiency cores are there if needed.

[the following commentary has been updated after using a Neo for a few weeks and finding out that it doesn’t struggle with multi core tasks as much as I expected, as evidenced here when stress testing. -May 2026, -JD]

Lets look into multi core, because compared to competitors the Apple multi-core situation is a different story than single store. With 6 cores (2 performance + 4 efficiency) versus 8-10 (or more) on competitors, the Neo’s 8,668 score is essentially M1-class. It actually trails Intel’s 8-core Ultra 5 226V (9,702) and the Snapdragon X Plus (11,345) in multi-threaded workloads. The M4 Air at 14,730 is 70% higher.

If you’re compiling a lot of code, running parallel builds, or doing sustained multi-threaded work a LOT, then this matters. For the Neo’s target audience, it likely won’t matter much b/c bursty workloads feel great and even with tons of multi-core and multitasking work it still manage to power through those moment. (Again granted, extreme encoding or long tasks are not going to be as fast as more-core competitors but I truly do no think the target audience is going to spend a large portion of their time on those tasks… if you need more than 6 cores then definitely get something else! For me, remember when the Intel Q6600 desktop CPU seemed SO fast and amazing to have quad core, I am pretty ecstatic with the Neo to get 6 cores in a lightweight portable package for ~$599.)

GPU performance at 31,286 (Metal) actually trails the M1 Air (33,148) slightly, despite the newer architecture. Five GPU cores versus the M1’s seven or eight means fewer parallel shader units. The M4 Air’s 54,630 is 75% higher. GPU-intensive work (video editing, 3D, gaming) is clearly not the Neo’s territory.

That 31,286 Metal number puts the Neo at M1-tier graphics. Cross-platform, two independent reviews (Notebookcheck’s MacBook Neo review and PCMag’s MacBook Neo review) put the Neo at 1,786 (PCMag 1,733) on 3DMark Steel Nomad Light and 3,999 (PCMag 3,947) on 3DMark Wild Life Extreme. That positions the A18 Pro 5-core GPU below the M4 Air (3,839 Steel Nomad Light / 8,735 Wild Life Extreme), well below the strongest ARM-Windows iGPU shipping today (Snapdragon X2 Elite Extreme in the Asus Zenbook A16 at $1,599 launch: 5,492 / 11,479), and below the AMD Radeon 860M tier (2,571 / 5,772). The Neo is closer to entry mobile iGPU territory than to Snapdragon X2. Intel’s newer Panther Lake Arc 140T/B390 leads the synthetic iGPU charts. Cyberpunk 2077 on the Neo (per PCMag’s MetalFX-enabled “For This Mac” auto preset) runs at 52 fps; at native 1080p Ultra it drops to 9 fps. The full GPU comparison post linked below has the rank order across every chip and explains the preset and source choices. The price axis matters as much as the benchmarks here:

Price (May 14, 2026)	Machine	Chip / GPU class	Thermal / memory context	Price basis
$599	MacBook Neo	Apple A18 Pro / 5-core Apple GPU	Fanless, 8 GB unified memory	Apple direct base
$899.99	Surface Laptop 13″	Snapdragon X Plus / Adreno X1	Fanless, 16 GB / 256 GB	Microsoft direct base config
$999	HP Omen 16″ (entry RTX 5060 laptop)	Discrete NVIDIA RTX 5060 Laptop	Actively cooled gaming laptop, 16 GB+ DDR5	Best Buy Q2 2026 promo example (typical non-promo retail in the $1,400-$1,600 range)
$999.99	Samsung Galaxy Book4 Edge 16″	Snapdragon X Elite first-gen / Adreno X1	Thin Copilot+ PC, 16 GB / 512 GB	Samsung / Best Buy current retail (down from a $1,450 launch list)
$1,099	HP EliteBook 845 G11	AMD Ryzen 7 8840U / Radeon 780M	Business-class 14″, 16 GB base config	Amazon base business listing; 32 GB / 512 GB business configs run $1,600-$1,700
$1,349.99	Asus Zenbook A14	Snapdragon X2 Elite X2E-88 / Adreno X2	Thin ARM ultraportable, base 16 GB / 512 GB	ASUS direct base (launched at $1,149.99 in April 2026 then adjusted upward post-launch; Best Buy’s higher-tier 32 GB / 1 TB row is currently around $1,699.99)
$1,699.99	Asus Zenbook A16	Snapdragon X2 Elite Extreme X2E-94 / Adreno X2	Actively cooled ultraportable, 48 GB / 1 TB	Best Buy current retail; launched at $1,599 in April 2026 with the 48 GB / 1 TB Windows 11 Home config

Representative US retail prices checked May 14, 2026. Listings move quickly: Surface, Samsung, and HP EliteBook are below where the original price-ladder chart placed them; the two Asus Zenbook X2 Elite parts are above their April launch prices; the HP Omen 16 is in the middle of a current Q2 promo cycle. The Neo is alone at $599; the cheapest comparison machine in this table at the moment is about 50% above it. This table is for price-tier context for a $599 fanless Mac, not a buying recommendation, and configurations and listings can change quickly.

UPDATE: Beyond Geekbench, More Benchmarks

Geekbench 6 is one useful datapoint but this next section digs a little deeper with sustained CPU performance under thermal load, compression and crypto throughput, integer and floating-point kernels, real-world video transcode. 

Before the tables, two notes. First, where the Neo is benchmarked from my own homelab, the testbed is the same throughout: MacBook Neo Mac17,5, A18 Pro 6-core, 8 GB unified, 256 GB SSD, macOS Tahoe 26.3.2, AC power, Low Power Mode off, Apple clang from the Xcode 16 command-line tools. The per-test manifest with start and end times, output files, and observed-state notes is at data/benchmarks/macbook-neo-hands-on/MANIFEST.md in the site repo. Second, where competitor numbers come from PCMag or Notebookcheck, the table caption says so explicitly. I do not combine measurements from different labs into a single blended row. 

Benchmark	Workload type	What it measures well	What it cannot tell you
Geekbench 6	Short, mixed real-app kernels	Burst CPU on PDF, HTML5, image, ML, ray tracer subtests; cross-platform	Sustained performance, GPU, memory pressure, specific apps, browser feel
Cinebench 2024	Cinema 4D float render, multi-minute run	Sustained multi-thread float throughput; cross-platform	Integer workloads, crypto, browser, anything outside Cinema 4D
HandBrake 1.8	Real video transcode using a fixed source file and preset	Real-world video transcode throughput when encoder, preset, and source file are identical	Comparisons across software vs hardware encode paths (VideoToolbox, QuickSync, NVENC), single-thread, GPU
7-Zip total MIPS	LZMA compression + decompression	Integer ALU and memory-subsystem throughput under multi-thread load	Float, GPU, real-app behavior
openssl speed rsa2048	Asymmetric crypto big-integer math (the openssl binary ships LibreSSL on macOS and OpenSSL on most Linux / Windows builds; both expose the same speed command)	Big-integer arithmetic throughput	Symmetric crypto (which uses hardware AES on Apple Silicon, Lunar Lake, etc.), memory bandwidth
C sieve 10M primes	Tight integer loop, single thread	Cache, branch predictor, integer ALU on a single thread	Anything threaded, I/O bound, or floating point
Matrix 512×512 GFLOPS (naive C)	Triple-nested float multiply-add, compiler auto-vectorization left on (Apple clang defaults)	Compiler vectorization quality and cache behavior on a naive kernel	Optimized BLAS / Accelerate / OpenBLAS results (an order of magnitude faster)
Speedometer 3, JetStream 2	Browser framework workload, JavaScript	Web-app responsiveness on the browser and engine used	Anything outside the browser; cross-OS comparisons are sensitive to browser version, JS engine, OS scheduler
5-minute all-core stress (this article)	Sustained all-core utilization	Thermal envelope, throttle onset, post-soak floor	Single-thread burst response, real-app interleaving

The thing this table makes obvious: every benchmark answers a narrow question. The Neo wins some of them and loses others by huge margins, depending on the workload.

The Neo across thermal states, across multiple benchmarks

Here is the same MacBook Neo running across cold, dev-workload, and post-thermal-soak states on every benchmark I have first-party data for. This is meant to make the burst-versus-sustained gap concrete on more than just Geekbench.

Protocol and what each thermal-state column means. “Cold pristine” applies only to the Geekbench 6 CPU rows (three runs after overnight rest, fan-assisted, two-minute cooldown between runs, three-run average reported as 3,569 / 8,879). The Geekbench Metal row is a single cool-state run on a separate morning, also fan-assisted. The C sieve and matrix “Cold pristine” cells are best-of-three runs from a “cold-ish” state in the same session (system was cool but not the overnight-rest cold of Geekbench, see MANIFEST). The OpenSSL speed rows are warm-state single runs measured roughly 20 minutes into the testing session and do not have a separate cold/post-soak comparison; a fresh cold-state OpenSSL pass is queued for the next lab run. “With Claude Code active” means Claude Code (Opus 4.6, 1M context window) idling in the foreground; this is descriptive of a developer’s typical background load, not a deterministic benchmark. “Post 5-min thermal soak” means the row’s benchmark was launched immediately after a 5-minute all-core stress command was stopped (CPU utilization peaked at 570% during the first 60 seconds of the stress run, dropped to about 200% as thermal throttling kicked in, and the post-soak benchmark was launched after the stress command exited). Where a cell shows a range (for example “39-41 ms” for the sieve under Claude Code), that range describes the three observed runs in that state rather than a best-of-three figure.

Test	Cold pristine	With Claude Code active	Post 5-min thermal soak	Source / notes
Geekbench 6 single-core	3,569	709 (-80%)	476 (-87%)	jdhodges.com lab; GB Browser runs 17527677, 17527733, 17527818 (3-run average)
Geekbench 6 multi-core	8,879	1,305 (-85%)	1,340 (-85%)	jdhodges.com lab
Geekbench 6 Metal (GPU)	31,275	(not run)	(not run)	jdhodges.com lab; Metal API, Apple platforms only
C sieve 10M primes (lower better)	18.8 ms	39-41 ms	124.2 ms (6.6x slower)	jdhodges.com lab; Apple clang -O2, single thread, auto-vectorize on
Matrix 512×512 GFLOPS (naive C)	15.9	7-12	3.85 (-76%)	jdhodges.com lab; naive triple-loop, Apple clang -O2 with auto-vectorization left on; not BLAS
OpenSSL speed rsa2048 sign/s	504	(warm state, separate run)	(not retested)	jdhodges.com lab; openssl speed rsa2048; the macOS openssl binary is LibreSSL 3.3.6
OpenSSL speed -evp sha256 @ 8K	2.1 GB/s	(warm state, separate run)	(not retested)	jdhodges.com lab; LibreSSL 3.3.6; SHA throughput here is implementation-dependent and almost certainly uses Apple Silicon SHA hardware instructions
7-Zip total MIPS	(pending fresh cold run)	(not run)	2,458 (provisional, throttled)	jdhodges.com lab; 7z b 6-thread, run started warm and spanned the thermal cliff. Cold-pristine 7-Zip is queued for the next lab session

The C sieve row is the one I want to call out. The same machine, the same binary, the same 10-million-element sieve, goes from 18.8 ms cold to 124.2 ms when launched immediately after the 5-minute stress test was stopped. That is a 6.6-times slowdown on a single-threaded integer workload that has nothing to do with Geekbench. The throttle is a property of the chassis and the fanless nature of the Neo, not the benchmark. If you only ever look at cold scores, you will never see this number.

The 7-Zip row is intentionally honest about a limitation in my own data: the run I have was started warm and ran straight through the thermal cliff, so its 2,458 MIPS total is a provisional, throttled figure, not a cold-pristine number. A clean cold-state 7-Zip run is on the list for the next lab session. Until that lands, the honest thing to do is publish what I have with the caveat, not omit the row.

Cinebench 2024 and HandBrake against the budget-laptop field

For a cross-laptop view that does not rely on Geekbench, PCMag’s MacBook Neo review (May 2026) ran the Neo against four PC competitors on Cinebench 2024 and HandBrake 1.8 on the same bench. All five rows below come from that single PCMag bench; the lab is consistent across the row. I have been a PC Magazine fan for decades and they continue to produce great reviews!

Machine (chip)	Cinebench 2024 SC	Cinebench 2024 MC	HandBrake 1.8 (lower better)
MacBook Neo (A18 Pro 6C, fanless)	129	308	11:20
MacBook Air 13″ 2025 (M4 10C, fanless)	168	849	5:07
Lenovo IdeaPad Slim 3x 15 (Snapdragon X)	96	687	5:17
Framework Laptop 12 (Intel Core)	95	293	14:26
Acer Aspire Go 15 (Intel N-series)	(not in PCMag CB chart)	(not in PCMag CB chart)	15:06

Source: PCMag MacBook Neo review, May 2026. Cinebench 2024 at default preset, HandBrake 1.8 with Fast 1080p30 preset, same PCMag test bench across the rows shown. Not every benchmark was reported by PCMag for every machine. HandBrake numbers are elapsed wall-clock time; PCMag does not explicitly identify whether hardware-accelerated encode paths (Apple VideoToolbox, Intel QuickSync) were disabled, so treat HandBrake elapsed times as a single-vendor, same-preset comparison rather than a strict “CPU encode only” comparison.

Two readings of this table matter. On single-core, the Neo’s 129 is the strongest result in the comparison after the M4 Air, ahead of both the Snapdragon X Lenovo and the Framework Intel system. On multi-core and HandBrake, the Neo loses decisively. The Snapdragon X Lenovo finishes the same HandBrake transcode in less than half the elapsed time, and PCMag’s reviewer’s word for it was that the Snapdragon X chip “outpaced the A18 Pro by a country mile.” This may be driven by the Neo’s thermal envelope, by different encoder paths between machines (Apple VideoToolbox availability versus Intel QuickSync versus pure software on each system was not separately reported by PCMag), or by both. The practical takeaway either way is that the Neo’s burst lead on single-core does not survive a long real-app transcode. The pattern is not Geekbench-specific.

Broader competitor context: Snapdragon X2, Lunar Lake, Ryzen

PCMag’s lineup leaves out the higher-end ARM-Windows, Lunar Lake, and Ryzen AI configurations that someone budget-shopping the Neo might be cross-shopping with at the next price tier up. Notebookcheck publishes chip-level Cinebench 2024 averages aggregated from device-level reviews. The numbers below are Cinebench 2024 only, because mixing in HandBrake or browser benchmarks across two different labs is exactly the kind of cross-source row I want to avoid. Treat each row as “how this chip scores on Notebookcheck’s chip page or in a specific Notebookcheck device review”, not as a direct head-to-head with the PCMag row above it.

Chip (representative laptop)	Cinebench 2024 SC	Cinebench 2024 MC	Source type and notes
MacBook Neo (A18 Pro 6C, fanless)	129	308	PCMag review result, $599 MSRP
Snapdragon X2 Elite Extreme X2E-94 (Asus Zenbook A16)	151	1,761	Notebookcheck Apr 2026 analysis review result, Performance mode, Zenbook A16 launched at ~$1,599
Snapdragon X2 Elite X2E-88 (Asus Zenbook A14)	149	1,471	Notebookcheck review result, Performance mode, Zenbook A14 launched at ~$1,149
Snapdragon X Elite X1E-84-100	127	866	Notebookcheck chip-page aggregate, first-gen X Elite
Snapdragon X Plus X1P-64-100	109	(not in our cited excerpt)	Notebookcheck chip-page aggregate, first-gen X Plus
Intel Core Ultra 5 226V (Lunar Lake, 8C)	112	572	Notebookcheck chip-page average. Tested laptops ran at 25 W, 30 W, or 35 W cTDP rather than Intel’s recommended 17 W, which favors the chip. SC range 111-114, MC range 530-609.
AMD Ryzen 7 8840U (Minisforum V3 convertible)	98	748	Notebookcheck single-sample, Minisforum V3 2-in-1 convertible (different thermal envelope than a typical 8840U ultrabook); read directionally only
AMD Ryzen AI 9 HX 370 (Strix Point, 12C)	117	1,085	Notebookcheck chip-page average across many laptop reviews, wide range of power limits and chassis; SC max ~121, MC max ~1,299; directional only

Sources: PCMag for the Neo row; Notebookcheck Snapdragon X2 analysis (April 7, 2026) for the Snapdragon rows; Notebookcheck Core Ultra 5 226V page; Notebookcheck Ryzen 7 8840U page; Notebookcheck Ryzen AI 9 HX 370 chip page. Different labs and different test laptops sit behind each row. The price tier on each row matters as much as the score, and the chassis (fanless vs actively cooled, ultrabook vs convertible vs mini-PC) matters at least as much as the chip.

The single-core picture is the most useful reading, with the cross-source caveat above kept in mind: in this mixed-source Cinebench 2024 context, the A18 Pro Neo’s 129 sits in the same band as the Lunar Lake Core Ultra 5 226V (112), the Ryzen 7 8840U datapoint shown (98), the first-gen Snapdragon X Plus (109), the first-gen Snapdragon X Elite X1E-84 (127), and the Ryzen AI 9 HX 370 chip-page average (117). Only the M4 (168) and the new Snapdragon X2 Elite parts (149-151) clearly lead it. For a $599 fanless laptop, “competitive with the burst single-core of nearly every chip in the chart except the M4 and the X2 Elite tier” is a strong result. Multi-core is the inverse story: the Neo’s 6-core 308 is well behind every actively-cooled competitor here because the 6-core fanless A18 Pro simply cannot stay near peak through a multi-minute Cinema 4D render. The cross-platform pattern matches what the Neo’s own thermal-cliff data shows.

Compression, crypto, and integer kernels: the Neo vs a $399 actively cooled Windows machine

Architecture: Is the A18 Pro Really “Just a Phone Chip”?

The internet discourse has centered on whether an “iPhone chip” belongs in a Mac. This framing is architecturally misleading.

The A18 Pro and M4 share the same DNA at the core level. Both are built on the ARMv9.2-A instruction set, both use Apple’s custom Everest performance cores and Sawtooth efficiency cores, and both are fabricated on TSMC’s N3E 3nm process. When you normalize Geekbench single-core scores by clock speed, you get approximately 857 points per GHz for both chips. The IPC (instructions per clock) is essentially identical.

They also share the same GPU shader core architecture (with hardware ray tracing and mesh shading) and the same 16-core Neural Engine rated at 35 TOPS. If Apple had branded the A18 Pro as “M4 Lite,” nobody would have blinked.

Where they genuinely differ is at the system level, and these differences matter:

Attribute	A18 Pro	M4
CPU cores	2P + 4E (6 total)	4P + 6E (10 total)
GPU cores	5	10
P-core clock	4.04 GHz	4.40 GHz
Memory bandwidth	60 GB/s	120 GB/s
System Level Cache	24 MB	16 MB*
Thermal envelope	~10W peak	~20-25W sustained
I/O	USB 3 + USB 2	Thunderbolt 4, PCIe 4.0

* Apple does not officially publish full cache hierarchies. SLC figures from cpu-monkey.com and nanoreview.net.

The memory bandwidth gap (60 vs 120 GB/s, a full 2x) is the most consequential difference. It limits any workload that’s memory-bound: large matrix operations, high-bitrate video encoding, GPU-intensive rendering. The A18 Pro partially compensates with a larger System Level Cache (24 MB vs the M4’s reported 16 MB), which reduces how often it needs to hit main memory.

The thermal difference also matters. The A18 Pro was designed for an iPhone, where sustained power is roughly 4W. The Neo’s larger fanless chassis allows somewhat more thermal headroom, but under prolonged multi-core loads it will throttle sooner than an M4 in a MacBook Air with its dedicated heatsink.

Bottom line: “Baby M4” is a useful shorthand. The shared core design means everyday responsiveness is M4-class. The system-level differences (bandwidth, thermals, I/O) mean sustained workloads are not.

Silicon Economics: How Apple Hits $599

The Neo’s price becomes less surprising when you understand the chip economics. The A18 Pro’s die measures approximately 105 mm² on TSMC N3E, confirmed by TechInsights die photography. That’s small. And small means cheap.

At 105 mm², the A18 Pro is 25% smaller than the M4 (~140 mm²) and 76% smaller than the M4 Max (~440 mm²). Smaller dies yield dramatically more chips per wafer and have higher yield rates because there’s less silicon for defects to land on.

A standard 300mm TSMC wafer produces approximately 586 gross dies at 105 mm² (per Arete Research). After 16 months of production maturity on N3E, yields are estimated at 85-90%, giving 498-527 good dies per wafer. At Apple’s estimated wafer cost of $18,000-$20,000 (per Ben Bajarin/Creative Strategies and Morgan Stanley), that works out to $34-40 per die before packaging and test. Fully loaded: roughly $38-47 per SoC.

Compare that to an M4 at ~140 mm² yielding approximately 430 gross dies, or an M4 Max at ~440 mm² yielding maybe 130. The A18 Pro costs Apple roughly one-third what an M4 costs and one-quarter what an M4 Max costs in raw silicon.

What’s even crazier?! Apple ships approximately 230 million iPhones annually. 🤯 The A18 Pro has been in volume production since September 2024. All mask costs ($10-20M for a 3nm EUV tapeout) and design engineering were amortized across hundreds of millions of units before a single Neo shipped. The marginal cost to Apple of routing A18 Pro dies into the Neo is wafer cost plus packaging. Zero incremental R&D.

The Neo may also absorb binned A18 Pro dies that failed their sixth GPU core during iPhone production. The Neo ships with 5 GPU cores; binned dies are perfectly suitable. This is standard industry practice (AMD and Nvidia do the same) and further improves Apple’s effective yield.

Adding up the full estimated BOM (SoC, memory, storage, display, chassis, battery, keyboard, wireless), the total lands at roughly $200-290. At $599 retail, that implies approximately 50-58% gross margin before R&D, marketing, and distribution. This is consistent with Apple’s company-wide gross margin of 47% on $436 billion in revenue. The Neo is not a loss leader. It’s a profitable product.

The 2026 RAM Shortage: Why 8GB Is Strategic, Not Just Cheap

The most common criticism of the MacBook Neo is the 8GB RAM ceiling with no upgrade path. Every Windows and Qualcomm competitor at this price ships with 16GB. Apple’s choice looks stingy until you understand the 2026 DRAM market.

The 2026 DRAM shortage is not a typical supply/demand cycle. It is a structural reallocation of global memory fabrication capacity toward AI infrastructure. The mechanism works like this:

High Bandwidth Memory (HBM), the DRAM used in AI accelerators like Nvidia’s H100/B200 GPUs, consumes approximately 3x the wafer area per gigabyte compared to standard DDR5 or LPDDR5x. This has been confirmed by Micron executives and independently by EE Times. HBM stacks require larger dies optimized for Through-Silicon Via interconnects, and yields for 12-high stacking run only 50-60%.

Samsung, SK Hynix, and Micron control 93% of global DRAM production. All three have aggressively reallocated capacity: up to 40% of advanced wafer output now goes to HBM. Micron exited the consumer memory market entirely in December 2025. As IDC put it: “Every wafer allocated to an HBM stack for an Nvidia GPU is a wafer denied to the LPDDR5X module of a mid-range smartphone or the SSD of a consumer laptop.”

The pricing tells the story. DDR5 32GB kits that cost $120 in Q3 2025 hit $350 by Q1 2026. Memory’s share of a PC’s bill of materials rose from 16% to 23% (Gartner). TrendForce projects a 90-95% quarter-over-quarter jump in PC DRAM contract prices for Q1 2026. Data centers will consume 70% of all memory chips manufactured in 2026.

The downstream effects are severe. Gartner projects global PC shipments will fall 10.4% in 2026, with average prices rising 17%. Lenovo, Dell, HP, Acer, and ASUS have confirmed 15-20% price hikes. Gartner’s projection: “The sub-$500 entry-level PC segment will disappear by 2028.”

This is the context that makes Apple’s 8GB choice strategically interesting, not just cheap:

Cost savings are real but not the whole story. At shortage pricing, 8GB of LPDDR5x costs Apple roughly $25-35. Doubling to 16GB would add $25-35 per unit. On a $599 product that’s significant, but Apple’s 47% gross margin on $436B revenue means an extra $30 is absorbable. This isn’t purely forced by economics.

The memory controller is a genuine constraint. The A18 Pro was designed for the iPhone 16 Pro, which has always shipped with 8GB. The LPDDR5x controller is configured for that package. Upgrading to 16GB would require different memory packaging and PCB routing. Apple could have designed around this, but chose not to for a first-generation budget product.

The shortage creates a pricing umbrella. As competitor laptop prices rise 15-20%, Apple’s fixed $599 becomes more competitive every month without Apple doing anything. A $600 Windows laptop that shipped with 16GB in mid-2025 now costs $700-750 for the same specs. Apple’s decision to halve the RAM halves its exposure to the shortage while competitors eat the full price increase.

The ecosystem math works. A Neo buyer who subscribes to iCloud+ and Apple One generates $240-480 in services revenue over a 2-year device lifecycle. At those numbers, the Neo’s hardware margin matters less than converting a Chromebook user into the Apple ecosystem.

Who Should (and Shouldn’t) Buy This

The MacBook Neo excels at: web browsing, email, document editing, streaming, messaging, light photo work, and running Apple Intelligence on-device. Single-core performance faster than any Mac until the M3 generation means everyday tasks will feel snappy.

It is not for: development work, gaming, content creation, video editing, virtual machines, heavy multitasking, or anything that regularly exceeds ~1.5-2GB of available application memory (after macOS overhead). The I/O is also a genuine limitation: one USB 2.0 port is one USB 3.0 and no Thunderbolt means a lot of limitations compared to systems with more (and faster [ports].

The ~$400-$500 gap to the MacBook Air is substantial, but so is what it buys: 2x RAM, 2x multi-core performance, Thunderbolt, MagSafe, backlit keyboard, P3 display, Wi-Fi 7, and a 12MP camera as well as more external displays being supported. If you can afford the Air, you may want to buy the Air. The Neo exists for people where ~$1,000 is not an option or they simply don’t want to spend that much.

Conclusion

The MacBook Neo is a genuinely impressive piece of engineering at an unprecedented Apple price point, strategically timed to exploit a market in crisis. The A18 Pro is not a compromise chip. It is the same core architecture as the M4, running at M3-to-M4 class performance for single-threaded work. Apple reused mature iPhone silicon at massive scale, eliminated incremental R&D cost, and shipped a product with healthy margins at $599.

The defining constraint is the 8GB memory ceiling and thermal issues, both those ceilings are product of engineering constraint (the A18 Pro’s memory controller), market economics (DRAM shortage pricing), and strategic calculation (ecosystem conversion at maximum volume), etc. It will age poorly, it is not upgradeable, and Apple knows this. The second-generation Neo with 12GB or 16GB is already the obvious product.

But right now, in March 2026, with sub-$500 PCs disappearing and average laptop prices climbing 17%, a $599 MacBook with M3-class single-core performance, an aluminum chassis, and 16 hours of battery life is the most interesting thing Apple has shipped in years. Not because it’s the best Mac. Because it’s the most strategically significant one.

Thanks for reading!

More MacBook Neo Coverage

• MacBook Neo Review: The $599 Mac That Benchmarked Itself: Full hands-on review with thermal testing, teardown photos, and the BeBox nostalgia trip.
• Can MacBook Neo Run Claude Code?: First-party resource usage data, the 80% Geekbench drop, and what throttled performance actually feels like in practice.
• MacBook Neo vs. Best Laptops 2026: How the Neo stacks up against Windows alternatives in every price tier.

If you enjoyed this deep dive, you might also like a blast from the past: the BeBox, a 1996 dual-PowerPC tower from Be Inc. It has some interesting parallels to the Neo, from the company which Apple considered buying before choosing NeXT! (but it has WAY more ports lol)

---

Sources: Apple Newsroom, MacRumors (March 5, 2026 benchmarks), Geekbench 6 Browser, TechInsights die analysis, Ben Bajarin/Creative Strategies (wafer cost analysis), Tom’s Hardware, IDC Global Memory Shortage Crisis Report (Feb 2026), Gartner PC Market Forecast, EE Times (“The Great Memory Stockpile,” Jan 2026), IEEE Spectrum, Network World, TrendForce, Counterpoint Research, AppleInsider, Six Colors.  PCMag MacBook Neo review (May 2026), Notebookcheck Snapdragon X2 analysis, Notebookcheck chip pages (Core Ultra 5 226V, Ryzen 7 8840U, Ryzen AI 9 HX 370), Primate Labs Geekbench 6 workload documentation, and first-party jdhodges.com lab runs documented in data/benchmarks/macbook-neo-hands-on/. Estimated figures are noted as such and rely on public industry analysis and standard semiconductor cost modeling.