Files
scrypted/plugins/objectdetector/src/performance-profile.ts
2023-06-27 23:23:02 -07:00

33 lines
1.2 KiB
TypeScript

import os from 'os';
let totalGigahertz = 0;
export function getMaxConcurrentObjectDetectionSessions() {
const cpus = os.cpus();
// apple silicon cpu.speed is incorrect, and can handle quite a bit due to
// gpu decode and neural core usage.
// .5 detect per cpu is a conservative guess. so an m1 ultra would handle 10
// simultaneous camera detections.
// apple silicon also reports cpu speed as 24 mhz, so the following code would
// fail anyways.
if (process.platform === 'darwin' && process.arch === 'arm64')
return cpus.length * .5;
let speed = 0;
for (const cpu of cpus) {
// can cpu speed be zero? is that a thing?
speed += cpu.speed || 600;
}
totalGigahertz = Math.max(speed, totalGigahertz);
// a wyse 5070 self reports in description as 1.5ghz and has 4 cores and can comfortably handle
// two 2k detections at the same time.
// the speed reported while detecting caps at 2500, presumably due to burst?
// the total mhz would be 10000 in this case.
// observed idle per cpu speed is 800.
// not sure how hyperthreading plays into this.
return Math.max(2, totalGigahertz / 4000);
}