How to tune fan curves when GPU runs at 100% but CPU idles for local inference
TechNerd10191 · reddit · 2026-09-24
The author is building a local inference workstation (9950X + RTX PRO 6000, North XL case, 7 Noctua fans, Fedora 44). The problem: during inference the CPU nearly idles while the GPU runs at 100%, so CPU-temperature-based fan curves leave fans slow exactly when the GPU needs airflow. Preferring not to run fans at 100% constantly, they ask multi-GPU workstation owners how to set fan curves based on GPU temperature instead.
More from Infra
- Dev Slams SGLang/vLLM Stacks as Buggy: Massive Low-Level Perf Work Remains — tekbog · 2026-09-24
- Anthropic Signs $517B in Compute Deals in Just 11 Months — DavidLinthicum · 2026-09-24
- Cohere expands Model Vault dedicated inference platform to Canada — cohere · 2026-09-24
- Cohere's Model Vault lands in Canada with single-tenancy and lower TCO — cohere · 2026-09-24
- System design is a quantitative engineering exercise: just build a spreadsheet and do the math — charles_irl · 2026-09-24
- Syscall Shims Like gVisor Deserve a Spot in the Sandbox Isolation Landscape, Argues Engineer — sloppenheimer · 2026-09-24