Sharing measurements in case it's useful to others running dedicated servers or just curious about the post-patch behavior of this game.
Context: Windrose launched on Steam EA two weeks ago. Pixel Operative, TechSpot and Tom's Hardware reported that it was writing ~108 GB/hr to disk continuously, idle or not, due to three stacked RocksDB instances with undersized memory caches. A patch (0.10.0.4, April 30) was deployed claiming to address it.
Setup:
- Dedicated server, DeploymentId 0.10.0.5.120-073042fb (post-patch)
- 2× NVMe in RAID 1 (Samsung MZVLB512HBJQ)
- No players connected during measurement window
- Measurement via `docker stats` (block I/O) and `nvme smart-log`
Post-patch measurements:
- 2-hour idle window: 69.6 GB written
- Sustained rate: ~35 GB/hr with zero players connected
- Annualized: ~307 TB/year from this one process
- Both NVMe drives now report Critical Warning 0x4 (NVM subsystem reliability degraded)
- Percentage Used: 122% and 128% after ~3 weeks of mixed workload (game + other production loads)
For reference:
- Consumer TLC NVMe endurance: 300–600 TBW
- Enterprise NVMe endurance: 600–1800 TBW
- Typical game write activity: a few hundred MB to ~1 GB/hr active, near-zero idle
So the patch reduced writes by 60–75% from the original 108 GB/hr baseline, but the absolute number is still 20–100× higher than a normally-behaved game, and the architectural cause (3× RocksDB with undersized caches) is unchanged. The two-week window before the patch is also unrecoverable wear for the ~1.5M players who installed during that period.
https://korben.info/ce-jeu-steam-flingue-votre-ssd-en-silence.html
https://www.techspot.com/news/112271-early-access-pirate-game-quietly-killing-ssd-without.html