Performance Discrepancy Obsidian

Steps to reproduce

  1. Clone a large Git-based vault (~12,000 markdown files with backlinks and frontmatter metadata).

  2. Open the vault on:

    • Machine A: MacBook Pro (M1 Pro, 64 GB RAM)
    • Machine B: Windows 10 Workstation with AMD Threadripper 3995X (64-core), 64 GB RAM
  3. I do not : Disable third-party plugins and use the default theme (sandbox mode).

  4. Launch Obsidian and start navigating the vault, using backlinks, graph view, and file search.

Did you follow the troubleshooting guide? [Y]

Expected result

Similar performance across both high-spec systems, with vault loading, intellisense indexing, and backlinks populating at comparable speeds — especially since Machine B is significantly more powerful in terms of core count and single-thread performance.

Actual result

  • MacBook Pro (M1 Pro) loads and renders Obsidian vault significantly faster (~5–10 seconds).
  • Windows 10 (AMD 3995X) takes significantly longer (30+ seconds), with a delay in loading backlinks, graph view, and internal link previews.
  • Heavy use of GPT-related extensions (code blocks + live completions) and graph connections may worsen performance on Windows but not on macOS.

Environment

  • Obsidian version: 1.5.12
  • Mac: macOS Ventura / Sonoma, 64 GB RAM, M1 Pro
  • PC: Windows 10 Pro, AMD Threadripper 3995X, 64 GB RAM
  • Vault size: ~12k Markdown files, versioned under Git
  • Plugins: (All disabled during test)
  • Theme: Default
  • Mode: Restricted Mode ON

Additional information

  • The Obsidian vault is heavily interlinked and uses a number of backlinks and internal references.
  • The graph view and connection load time is much slower on Windows.
  • Disabling GPT and intellisense-related extensions helps slightly on Windows, but macOS still consistently performs better.
  • Suggests possible OS-level optimizations or differences in how file I/O and rendering threads are managed between macOS and Windows.

There are no OS-level performance optimization done by us. Perhaps, electron/chrome does something but I am not aware of it and we would have no control over it.

Another simple explanation is that the M1 has much higher single threaded performance than the 3995X.

I am gonna move this thread to the help section because it doesn’t meet our BR criteria nor I think there’s a bug here.