useCPU

Source: packages/sdk/src/hardware/useCPU.ts

Reads the CPU topology from /proc/cpuinfo (per-core Arm part ids mapped to names such as Arm C1-Ultra, Arm C1-Pro) and cpufreq sysfs (cpuinfo_max_freq, scaling_cur_freq, scaling_governor). Android does not expose whole-system /proc/stat to apps, so the two load signals are (a) cluster frequency utilisation and (b) this app’s own CPU share.

Verified on Pixel 11 Pro: 1x Arm C1-Ultra @ 4.11 GHz + 4x Arm C1-Pro @ 3.38 GHz + 2x Arm C1-Pro @ 2.65 GHz, governor sched_pixel. Topology is read once on mount; load polls every 1,000 ms until unmount.

Signature

function useCPU(): {
 coreTopology: string;
 coreCount: number;
 cpuLoadPercent: number | null;
 appCpuPercent: number | null;
 cores: CoreFrequency[];
 clusters: { part: string | null; name: string | null; maxMHz: number | null; count: number }[];
 governorMode: string;
 lastBenchmarkDurationMs: number | null;
 isBenchmarking: boolean;
 benchmarkCPU: () => Promise<number>;
 source: TelemetrySource;
};

Outputs

Field	Type	Description
coreTopology	string	Human-readable cluster summary built from real cluster data, e.g. "1x Arm C1-Ultra @ 4.11 GHz + 4x Arm C1-Pro @ 3.38 GHz". "unknown" before the first read.
coreCount	number	Cores visible to this process. 0 until the native module answers; 7 on the Tensor G6.
cpuLoadPercent	number | null	Cluster frequency utilisation in percent: current ÷ maximum clock averaged over cores. Not scheduler load. null when the cpufreq sysfs files are unreadable.
appCpuPercent	number | null	This process’s CPU time as a percentage of all cores, from Process.getElapsedCpuTime over wall time. null on the first sample because two readings are needed. Recorded with provenance derived.
cores	{ index: number; part: string | null; name: string | null; curMHz: number | null; maxMHz: number | null; minMHz: number | null }[]	Per-core detail: kernel index, Arm part id, mapped name, and current/max/min clock in MHz. Empty before the first read.
clusters	{ part: string | null; name: string | null; maxMHz: number | null; count: number }[]	Cores grouped by part id, which is how you tell prime, performance and efficiency cores apart.
governorMode	string	Kernel cpufreq governor for cpu0, "sched_pixel" on this device. "unknown" before the read; not settable without root.
lastBenchmarkDurationMs	number | null	Duration in milliseconds of the last benchmarkCPU() run. null until one has run.
isBenchmarking	boolean	true while the benchmark occupies the JS thread. Use it to disable the trigger control.
source	TelemetrySource	'hardware' when the PixelNative module is present, 'unavailable' otherwise.

Functions

Function	Inputs	Returns	Description
benchmarkCPU()	none	Promise<number> — run duration in milliseconds, also written to lastBenchmarkDurationMs	Runs a real single-threaded prime sieve (trial division to 60,000) on the JS thread. It measures Hermes single-thread throughput, not the system, and blocks the UI while it runs.

Example

const { coreTopology, cpuLoadPercent, cores, benchmarkCPU } = useCPU();
<Text>{coreTopology}</Text>
<Text>{cpuLoadPercent ?? '—'}% · {cores.map(c => c.curMHz).join('/')} MHz</Text>