Leaked Apple M5 9 core Geekbench scores

Original Article ↗ Hacker News Discussion ↗

M5 vs M4 Performance Uplift

  • Leaked 9‑core M5 iPad Pro shows ~10–12% single‑core and ~15–16% multi‑core uplift over 9‑core M4 at the same max clock, with more L2 cache and 12 GB RAM baseline.
  • GPU uplift is discussed as ~30–40%, consistent with recent A‑series gains, and seen as the most significant part of this generation.
  • Some extrapolate MacBook single‑core to ~4,300–4,400 Geekbench, continuing the steady M1→M5 curve.

Benchmarking Nuances (Geekbench, SME/AMX)

  • Debate over Geekbench 6: part of recent jumps come from new SME support; some argue this overstates “real” IPC gains, others report similar proportional wins in real builds.
  • Confusion/clarification around AMX vs SME: both are Apple matrix engines, SME is newer and more directly usable; some say “no apps use it”, others note any Accelerate‑using or LLVM‑17‑compiled code can.
  • Broader argument about general‑purpose benchmarks vs real workloads; Geekbench’s crowd‑sourced data is polluted by VMs and misconfigured systems.

Core Counts, Process Nodes, and Architecture

  • Base M‑series core mixes are listed (M1–M5, iPad vs Mac); M4/M5 emphasize more efficiency cores.
  • Some ask why 9 cores; answers include binning (one core disabled) and that non‑power‑of‑two counts aren’t unusual.
  • M5 is believed to be on TSMC N3P, not 2 nm; discussion that “nm” names no longer map to real dimensions.

GPU, AI, and Local ML Workloads

  • Several care more about GPU and AI accelerators than CPU: matmul units, neural accelerators, and GPU throughput matter for LLM and diffusion workloads.
  • Apple Silicon is seen as far behind consumer Nvidia GPUs for heavy AI, but good for “decent” local inference on laptops and phones.
  • Some imagine Apple could challenge Nvidia on client‑side inference if they prioritized AI‑friendly GPUs/NPUs and software; others say Nvidia’s datacenter stack remains untouchable.

Real‑World Use, Upgrade Cycles, and M1 Longevity

  • Many commenters on M1/M1 Pro/Max say they still “feel fast” 4–5 years on; most don’t see a strong need to upgrade for everyday dev or office work.
  • Exceptions: heavy Rust/C++ builds, big CFD workloads, and local LLMs benefit meaningfully from newer chips.
  • Several users regret low‑RAM configs more than older CPUs; resource‑hungry workflows (Chrome, Slack, Docker, IDEs) strain 8–16 GB.

iPadOS Constraints vs Mac macOS

  • Strong theme: iPad hardware is “massively overpowered” for what iPadOS allows; users can’t exploit the SoC like on macOS (limited sideloading, no real terminals, blocked hypervisors, app‑store gating).
  • Others counter that creative apps (CAD, DAWs, sculpting, video, Logic/Final Cut on iPad) do push the hardware, especially with new multitasking in iPadOS 26.
  • Frustration that iPad Pro can’t simply run macOS or mac‑class apps, despite near‑identical SoCs.

RAM, Storage, and ‘8 GB’ Debate

  • Some call continued 8 GB base configs “borderline unethical” for longevity; others report 8 GB M‑series working fine for light to moderate use.
  • Real‑world complaints: 8–16 GB machines hitting swap under Chrome + comms apps + dev tools; slowdowns tied more to RAM than CPU.
  • 256 GB base SSD and limited ports on Airs are seen as major constraints by some, irrelevant by others who rely on docks and external drives.

Apple Silicon vs x86 and Snapdragon

  • Consensus that Apple leads in single‑core perf and perf/W in laptops; AMD seen as competitive on desktops, weaker on mobile efficiency.
  • Snapdragon X‑series is noted as “close” in some benchmarks (and improving), but still typically behind in efficiency and mac‑class system integration.
  • Debate over which benchmarks to trust (Geekbench vs Cinebench vs PassMark), and how much process node advantage vs Apple design explains the gap.

Openness, Linux, and Asahi

  • Many lament locked‑down firmware, closed drivers, and difficulty running Linux on M‑series (especially iPads); some wish Apple would officially support Linux use.
  • Asahi Linux is praised; recent focus is on upstreaming existing M1/M2 work before tackling newer SoCs. Loss of key contributors (especially GPU devs) is seen as a serious blow by some, “project basically completed for those gens” by others.
  • Snapdragon ARM laptops + Linux are desired by some, but commenters warn about ARM PC/Linux ecosystem fragmentation and weak vendor support.

Future Products: Touch Mac, Mac Pro, ARM MacBook Lite

  • Rumors cited of:
    • Touch‑enabled OLED MacBook Pro around M6 timeframe.
    • A cheaper “MacBook” using an A‑series (iPhone‑class) chip.
    • Repositioned Mac Pro/Studio as high‑RAM AI/ML workstations, possibly with M5 Ultra/Extreme.
  • Opinions on touch Mac are mixed: some want it for stylus and presentations; others fear worse anti‑glare and accidental touches.