Summary of "Процессоры Apple M — всё. RISC был оправдан?"

Summary — technological analysis, product implications, and recommendations

High-level thesis

Apple Silicon (ARM-based Apple M series) succeeded because of three pillars:
- ARM ISA energy efficiency.
- Deep SoC integration (CPU + GPU + NPU + media engines + unified memory).
- Full vertical control (hardware + OS + apps).
The M1 launch (2020) was a watershed: excellent battery life, strong single-chip performance, and broad device applicability (Air, mini, iMac, iPad).
Since then, per‑generation CPU/GPU gains have often looked modest (roughly ~10% per core in many steps). External competition (notably Qualcomm’s Snapdragon X Elite) and limits in TSMC process scaling have created more uncertainty about Apple’s future advantage.

Technical details and manufacturing analysis

ARM vs x86

ARM (RISC) benefits from fixed-length instructions and simpler pipelines, which generally yield better energy efficiency compared to x86’s legacy complexity — a key reason Apple moved to ARM for its Macs.

SoC integration

Apple’s advantage is packing CPU, GPU, NPU, media engines, and unified memory into one optimized chip tailored to macOS/iPadOS, enabling efficiency and performance gains that are hard to match with discrete components.

TSMC process nodes (brief)

N5 (5 nm): used for M1.
N5P (improved 5 nm): used for M2 (minor transistor tweaks + better efficiency).
N3 family (3 nm variants): includes N3b, N3, N3e, N3p, N3x. Variants differ in layer count, yield/defect characteristics, and compatibility; N3 was presented as a simplified/cheaper variant vs N3b.

Practical scaling limits observed

M2 increased transistor count (e.g., ~16B → ~20B) and used N5P for modest per‑core gains (~7–10%) with larger multithread gains by adding cores.
M3 was marketed as “3 nm” but real density gains over M2 were smaller than theoretical expectations because many chip components (SRAM, analog, IO) don’t scale as well. Measured uplift resembled prior generation steps (~10% per core, ~20% multithread).
M3 appears to have been fabricated across mixed N3-family choices (tradeoffs in density, yield, schedules), which helps explain why M3 didn’t deliver dramatic theoretical density improvements.

The M4 surprise

M4 arrived ≈190 days after M3. Early benchmarks (Geekbench cited) show large single- and multi-core gains (host claims ~25% overall; single-core near 3800; multi-core ≈14,600).
Plausible explanation: Apple used different N3-family variants and introduced GPU/CPU architectural changes. Apple prioritized getting a higher‑performing product quickly by splitting work across nodes and reusing/retargeting designs.

Ultra / multi-die and UltraFusion

M Ultra is two Max dies joined by an UltraFusion bus (the bus occupies significant area).
The M3 Max MacBook Pro omitted the UltraFusion bus — implying Apple did not intend to make an M3 Ultra, and planned to reserve Ultra-class chips for M4 or later.

Neural Engine (NPU) and numeric formats

Apple has reported NPU performance in different numeric formats (FP16 historically; INT8 more recently). INT8 yields higher TOPS numbers for the same silicon compared with FP16.
Claimed numbers (from the video/insiders):
- M4 ≈ 38 TOPS (INT8)
- M3 ≈ 18 TOPS (FP16)
- Snapdragon X Elite ≈ 45 TOPS (INT8)
Comparisons require caution: TOPS alone don’t represent real-world NPU performance — task type, memory bandwidth, software stack, and numeric format matter.

Competition and software ecosystem

Qualcomm Snapdragon X Elite (≈4 nm) is a credible competitor for Windows laptops. Highlights:
- 12 performance cores, Arena GPU, Exagon NPU.
- Wi‑Fi 7, support for up to 64 GB LPDDR5, Snapdragon X65 5G, NVMe via PCIe.
Software porting and emulation:
- Apple: Rosetta 2 allows x86 apps to run on Apple Silicon (often with ~10% performance loss for some workloads).
- Microsoft: Prism translator and first‑party efforts enable x86 apps on Windows-on-ARM; major vendors (Chrome, Spotify, Adobe) are porting apps to native ARM builds or intermediate representations.
Neural engines will strongly influence future workloads (AI-accelerated editing, rendering, plugins for Premiere/After Effects, DaVinci Resolve, Blender, etc.). TOPS counts are increasingly relevant but must be evaluated with real workloads.

Product and buying recommendations

Apple’s launch cadence and TSMC node complexities explain why some devices retained M2 while others received M4 or M2 instead of M3 (e.g., iPad Pro, Vision Pro).
The M3 Max MacBook Pro lacks the UltraFusion bus — so M3 Ultra is unlikely; high-end desktop Macs (Mac Studio, Mac Pro) are expected to skip M3 Ultra and move to M4 Ultra when available.
Practical advice (from the video):
- If you need a top desktop/workstation now, wait for M4‑equipped Macs (M4 Ultra) rather than buying a current M2 Ultra machine.
- If you already have an M1 Pro/Max, an upgrade depends on needs: upgrade now only if you require more power; otherwise, consider waiting for M4 Pro/Max in MacBook lines (likely in autumn).
- Expect WWDC to focus on desktop Macs with M4 variants; MacBook M4 Pro/Max models could appear in the fall.

Benchmarks and review guidance

Look for reviews and tests that cover:
- Real application performance: video/photo rendering and AI-accelerated workflows.
- Thermal behavior and sustained performance (throttling/noise).
- Synthetic benchmarks (Geekbench single- and multi-core) as indicators, but not as the only measure.
- NPU/AI tests that run real workloads rather than relying solely on synthetic TOPS figures.
Suggested comparisons to follow:
- M1 / M2 / M3 / M4 vs Snapdragon X Elite laptops (CPU, GPU, NPU, battery, thermals).
- Application-specific tests (Adobe suite, DaVinci Resolve, Premiere, After Effects, Blender) that exploit NPUs and media engines.

Sponsor / ad notes

The referenced video contained promoted mentions:
- MTS ver (a browser-based virtual world).
- An ad for realme 12+ 5G smartphone (Dimensity 7050, Sony Lyt 600 camera, 6.67” AMOLED 120 Hz, up to 2,000 nits, storage/RAM options: 8/256 or 12/512). These are peripheral to the Apple/TSMC/Qualcomm analysis.

Conclusions from the video

Apple’s transition to Apple Silicon was justified and transformational (M1).
Apple now faces manufacturing scaling limits and stronger Qualcomm competition; future advantages depend on node progress, NPU development, and sustained software ecosystem momentum.
The rapid arrival of M4 after M3 can be explained by TSMC node choices and Apple’s product sequencing across N3 variants, rather than implying a secret sudden leap.
Practical takeaway: wait for M4‑equipped Macs for top desktop/workstation needs; for portable upgrades, evaluate specific M3 vs M4 MacBook models and your workflow (video editors should pay particular attention to thermals and NPU/accelerator usage).

Main speakers / sources cited

Video narrator / host (unnamed).
Snas Laps channel (analysis referenced).
Mark Gurman.
MacTec channel.
TSMC (N5, N5P, N3-family descriptions).
Qualcomm (Snapdragon X Elite specifications).
Software/tech entities mentioned: Apple, Microsoft (Prism, Windows on ARM), Google, Adobe, Spotify, and others porting apps.

Note: Many numeric claims (benchmarks, TOPS, node names) came from insider reports and community calculations in the video and include uncertainty or assumptions noted by the narrator. Treat raw numbers as indicative rather than definitive.