2 ms·
Power on Tap: Dynamic control of P cores in M1 chips
- jeffbee 4y ago70ms to reach full speed seems extreme. Is it possible that the reported clock rate is being smoothed by some filter? An Intel Skylake CPU makes P-state transitions in a few hundred microseconds at most.
- chaboud 4y agoThis would be akin to the rate as set by the governor in Linux, which can be configured to respond somewhat slowly to rising demands. Server configurations of the ondemand governor in Linux often take 50-90ms to reach max state from idle in response to load, and other variants (e.g., interactive) cover over this by having features for user interactivity (common in Android devices). 70ms to max out sounds pretty reasonable to me, but I'd be very curious to see if Apple has added some interactivity boosts.
- KerrAvon 4y agoI would caution that this fellow tends to extrapolate conclusions from limited data. Take with a grain of salt unless confirmed by an Apple engineer.
- Maursault 4y ago> E cores are normally managed by macOS to run at frequencies of 600 MHz (idle), 972 MHz (one or few threads) or 2064 MHz (multiple threads). I'm confused by this statement. My understanding was the M1 Macs' thread-count matched the total number of cores, which (I thought, obviously,) means the M1 cores are not multithreaded.
- wtallis 4y agoThis seems to be referring to how many software threads are spawned by the benchmark and assigned to be scheduled on E cores due to their QoS settings.