Thermal management

• Sustained workload budgeting. The system uses CPU and GPU levels to keep the device inside a steady-state thermal envelope. Choosing a lower sustained level when you can leaves headroom for short bursts.
• Throttling. When sensors approach a threshold, the system reduces clock speeds. Your app keeps running, but per-frame cost rises and frame timing becomes more variable.
• Frame-timing adaptation. The runtime’s frame timing algorithm, FrameSync, is designed to absorb the variable frame cost that throttling produces — sliding-window statistics and outlier rejection keep frames landing in their target vertical sync intervals when workload spikes.
• Cool-down state. If sustained workload pushes temperature past a hard limit, Horizon OS may force the device into a low-power cool-down state until sensors recover. The user sees a cool-down notification.

• Build for the long session, not the demo. Apps often look fine for the first three minutes and then visibly degrade as the device heats up. Profile sustained sessions, not just startup.
• Stay below the sustained CPU and GPU level you actually need.Boosting is for short bursts (loading, scene transitions, physics-heavy moments). Sustained boost defeats the purpose and costs you battery and thermal headroom.
• Stagger expensive work across frames. Single-frame spikes are absorbed by FrameSync’s outlier rejection, but predictable per-frame patterns produce the smoothest output and the lowest sustained heat.
• Detect and adapt to system signals. Apps should detect battery saver mode and reduce visual fidelity or disable non-essential effects when the device asks for help.
• Test on the device, not the simulator. Thermal behavior is hardware-dependent — only on-device profiling reveals the real envelope.

• High shader complexity combined with high-resolution post-processing
• Continuous full-resolution passthrough alongside a heavy 3D scene
• Always-on physics simulation, particle systems, or dynamic lighting
• Failing to take advantage of fixed foveated rendering or App Spacewarp when the workload allows

• Performance and hardware — the broader hardware-resource model
• CPU and GPU levels — how the system selects clock speeds
• Boosting CPU and GPU levels — short-burst escalation
• Battery saver mode — how the system reduces performance under low battery
• FrameSync — how the runtime adapts frame timing under variable load