It is worth it to buy the fast CPU

Original Article ↗ Hacker News Discussion ↗

Apple vs x86, Laptops vs Desktops

  • Some argue “just use a Mac”: excellent perf/W, strong unified memory for local ML, great dev laptops.
  • Pushback: poor perf/$, limited cores vs high‑end x86, weak extensibility (PCIe, storage upgrades), flaky multi‑monitor, locked‑down OS and notarization overhead.
  • Consensus: Apple makes very good laptops, but not undisputed “killer dev machines,” and they don’t replace high‑core x86 workstations.

Where CPU Speed Actually Helps

  • Big wins: large C/C++ builds, Rust builds, linkers (with parallel linkers like mold/lld), LSP responsiveness, and heavy test suites.
  • Many report near‑linear scale with core count for compiles—up to memory bandwidth limits. Others see diminishing or negative returns beyond ~40 cores.
  • Faster SSDs and more RAM are repeatedly mentioned as equally or more important than raw CPU.

When Faster CPUs Don’t Help

  • Bottlenecks often elsewhere:
    • IO (especially NTFS + Defender on Windows), network/VPN, cloud APIs, security tooling (MFA, PIM), SaaS tools, slow ERP/PLM systems.
    • Single‑threaded or startup‑bound apps (e.g., Teams, some Rails/Angular/Java stacks).
  • For many, the longest waits are now CI pipelines or remote services, not local compiles.

Cloud Workstations and Remote Builds

  • Several big orgs use VDI/remote dev hosts or Bazel/Buck build farms: laptop becomes a thin client, heavy work is remote.
  • Experiences split: some find latency acceptable (<100 ms) and love the flexibility; others hate any added lag and distrust “cloud for everything,” especially for non‑web or graphics‑heavy work.

Economics of Developer Hardware & Corporate Policy

  • Strong sentiment that companies under‑invest in dev machines and ergonomics despite high dev salaries; penny‑wise, pound‑foolish.
  • Others highlight abuse: max‑spec laptops for light workloads, luxury chairs/desks, food perks gamed into “soft expense fraud.” This drives stricter controls and standardized mid‑tier configs.
  • A recurring view: top‑spec machines pay off quickly for high‑paid engineers with heavy local workloads; less clear for lighter or fully‑remote workflows.

Upgrade Cadence, Generational Gains, and “Good Enough”

  • Disagreement on progress: some say single‑core has roughly doubled/tripled in a decade; others note only ~7–13%/year, making frequent upgrades marginal.
  • Many 2020‑era CPUs (e.g., 5800X/5950X, older Threadrippers) are still “fast enough”; moving from “good” to latest often yields modest gains unless you were badly under‑spec’d.
  • Several anecdotes of decade‑old desktops still fine for typical dev/web use once given SSDs and more RAM.

Software Bloat vs Faster Hardware

  • A strong faction argues slow experiences are mostly software/architecture problems, not hardware; they advocate testing on low‑end/older hardware and poorer networks.
  • Counterpoint: forcing devs to work on slow machines just burns time and flow; better to develop on fast boxes and explicitly benchmark on constrained targets.