Alleged M4 MacBook Pro unboxing video

Original Article ↗ Hacker News Discussion ↗

Performance Gains M1 → M4

  • Users compute single-core Geekbench jumps: ~13% (M1→M2), ~19% (M2→M3), ~25% (M3→M4, if leak is legit).
  • Some attribute gains partly to new ARM extensions (SVE/SME) and higher clocks; others question how long frequency boosting can continue.
  • Debate over benchmarks: Geekbench seen as too short to show throttling; Cinebench and longer tests suggested as more relevant.
  • M4 iPad single-core is said to approach high-end desktop CPUs, but duration of peak performance is questioned.

Process Node & Architecture

  • Each M generation has coincided with new TSMC process (N5 → N5P → N3B → N3E).
  • N3E is described as slightly less efficient than N3B but with better yield; unclear how much that affects retail performance.
  • Some see Apple’s ARM push and AWS/Ampere/Raspberry Pi as having legitimized ARM more broadly; others emphasize pre-M1 server work.
  • Intel’s upcoming Lunar Lake (on TSMC) cited as evidence that Intel’s issues were more fab-related than ISA-related.

Upgrade Value & Real-World Reports

  • Many users with M1 Pro/Max or M1 Air report M3 as faster but not life-changing; M1 already a huge jump from Intel.
  • Common theme: if your M1 still feels sufficient, upgrade is “nice to have,” not necessary.
  • Gains from more RAM (e.g., 16→36 GB) can matter more than CPU gen for some workflows.
  • Some would upgrade mainly for features (MagSafe, better display, future OLED) rather than raw performance.

RAM, Bandwidth, and LLM / Heavy Workloads

  • Rumored 16 GB base RAM for all Macs seen as important for local LLMs.
  • Interest in max RAM (e.g., 192–256 GB) and memory bandwidth for large models and distributed systems.
  • Debate on comparing unified memory to GPU VRAM: disagreement over how to reason about bandwidth across multiple GPUs and parallelism modes.

I/O: Thunderbolt, TB5, MagSafe, USB‑C

  • Some disappointed by apparent lack of Thunderbolt 5; 40 Gbit/s seen as limiting for high-end SSDs, networking, and (hypothetical) eGPUs.
  • Others argue 40 Gbit/s is plenty for most and note real-world TB4 throughput is lower due to video reservation and protocol overhead.
  • Discussion of optical interconnects as a likely future for higher bandwidth.
  • MagSafe vs USB‑C:
    • Pro‑MagSafe: safer disconnects, preserves USB‑C ports for data, user preference.
    • Anti‑MagSafe: extra proprietary cable, USB‑C seen as sturdy enough and more universal.
    • Mixed experiences on port durability and dust/grit issues.

Core Scheduling & Docker

  • Some developers see erratic benchmark results attributed to tasks being scheduled on efficiency instead of performance cores.
  • Suggested workaround: taskpolicy to bias processes toward performance cores, though Docker integration is unclear.
  • More invasive workaround (disabling efficiency cores via custom kernel) is considered impractical on company machines.

Authenticity of the Leak & Supply Chain / Sanctions

  • Skepticism around the unboxing: box art matches earlier M3 designs, and Apple usually updates wallpapers per generation.
  • Speculation that this could be reused art, unofficial packaging, or not genuine at all; overall authenticity remains unclear.
  • Questions on how a Russian YouTuber obtained the device; some link it to Chinese manufacturing leaks and broader sanction evasion.
  • Side debate on sanctions:
    • One view: sanctions mostly hurt ordinary people, rarely elites; framed as ineffective and harmful.
    • Counterview: they do cost elites money but may still have limited geopolitical effect; cited as overused because alternatives are hard.