Skip to content

Linux vs Windows: 10–40% faster processing

  • Status: unverified
  • Applies to: Metashape Pro 2.x ; Metashape Standard 2.x
  • Edition: Pro / Standard
  • Diátaxis: explanation
  • Confidence: medium
  • Last reviewed: 2026-05-27

Confidence: medium. Based on community benchmarks (3 users, 3 hardware configurations, December 2024). Agisoft has not confirmed or denied the findings. No official documentation addresses OS-level performance differences. This article cannot be demonstrated via a sample dataset — it requires comparative benchmarking on identical hardware under two operating systems.

The finding

Multiple independent benchmarks (December 2024, Metashape 2.1.3+) report that Metashape runs 10–40% faster on Linux than on Windows on identical hardware.

Hardware Linux (Rocky 9.5) Windows 11 Δ
AMD Ryzen 9 9900X + RTX 4090 4380 s 6016 s +37% slower on Win
Intel i9-9900K + RTX 2080 Super 8999 s 11815 s +31% slower on Win

Source: Agisoft forum, 2024-12-07, Metashape 2.1 (permalink)

Controlled isolation

A controlled test with locked CPU (4.4 GHz) and GPU (1700 MHz) frequencies on debloated Windows 10 LTSC vs Arch Linux showed a narrower gap: ~10% difference.

The residual ~5% is attributed to OS overhead:

Metric Windows Linux
Processes ~141 ~62
Threads ~1720 ~149
OS handles / kernel threads ~50,500 ~259

Root causes

The performance gap has multiple contributing factors:

  1. Windows background services. Telemetry, Defender real-time scanning, Superfetch, Cortana, Windows Update. Each consumes CPU cycles and causes context switches. Disabling these narrows the gap significantly.

  2. Thread scheduling overhead. The Windows kernel maintains far more threads and handles than a minimal Linux installation. Each context switch has a cost; with 1700 threads vs 149, the aggregate overhead is measurable on compute-heavy workloads.

  3. WDDM driver model. On Windows, NVIDIA GPUs use the Windows Display Driver Model (WDDM), which adds overhead for GPU command submission compared to Linux's direct kernel-mode driver. This affects all GPU-accelerated stages (feature detection, depth maps).

  4. File system. NTFS vs ext4/XFS may contribute for I/O-heavy stages (loading images, writing depth maps). Not conclusively isolated in these benchmarks.

What Agisoft says

Agisoft support requested detailed per-step logs for investigation but made no official statement on the performance difference as of December 2024. In an older thread (2017), Agisoft support explicitly recommended trying Linux for multi-GPU setups where Windows showed low utilization:

"I can suggest to try running PhotoScan on Linux [...] As there shouldn't be an issue related to WDDM driver, I believe the utilization of the cards would be considerably better." — Alexey Pasumansky, 2017-05-04, PhotoScan 1.3 (permalink)

Practical recommendations

  • If you process large datasets regularly, the 10-40% speedup on Linux compounds into significant time savings. A dual-boot or dedicated Linux processing machine pays for itself quickly.

  • If you must stay on Windows, debloat aggressively: disable Defender real-time scanning during processing, disable telemetry, Superfetch, and unnecessary services. Windows 10 LTSC is closer to Linux performance than stock Windows 11.

  • The difference is per-step, not per-stage. Every processing step (match photos, align, depth maps, mesh, texture) is affected, not just GPU-heavy stages. This suggests the overhead is systemic (OS scheduling + driver model), not specific to one algorithm.

  • SLI/NVLink does not help. Multi-GPU setups benefit from Linux even more because WDDM overhead affects each GPU independently. See Multi-GPU setups and GPU usage by stage for details.

Caveats

  • Sample size is small (3 users, 3 hardware configurations).
  • The 30-40% figure is for "stock" Windows vs Linux. With debloating, the gap narrows to ~5-10%.
  • Agisoft has not confirmed or explained the difference officially.
  • Results are from December 2024 (Metashape 2.1.3); newer versions may differ.
  • Linux requires manual NVIDIA driver installation and lacks the Metashape GUI's full feature parity (some dialogs differ slightly).

References

  • Metashape Pro User Manual (2.3), § System Requirements — lists Windows, macOS, and Linux as supported platforms without noting performance differences.
  • Metashape Python API Reference (2.3.1): Application.gpu_mask, Application.enumGPUDevices().