On 01/ 4/11 03:44 AM, Milan Jurik - Oracle Czech wrote:
With close to zero activity in userspace, something in kernel must consume 1/4 of CPU. Or could prstat miss some spikes of activity in userland, which are pushing CPU to the highest P- and C-state?
I doubt there is consumption of 1/4 of a CPU (i7 980X, 6C/12T). At best we can say is powertop showed the CPU stayed in the highest P-state 1/4 of the time.
Still, is powertop on b154 looking similar in amount of time spend in P- and C-states? Didn't nVidia drivers change between builds? Are these drivers pushing GPU to some sleep mode?
The drivers didn't change between 154->155. I changed the NVIDIA driver a long time ago to normally wake up only 4 times per second. Before taking the powertop snapshot I had verified the GPU had already transitioned into it's lowest power state and had not moved for over a minute so there shouldn't have been any device interrupts for dynamic power management. FWIW, the ahci#0 interrupt is firing much more frequently. For 1 second intervals, intrstat shows anywhere between 0 - 270 interrrupts per second (compared a near constant 4 for the NVIDIA device). If I look closely, the ahci peak matches the spikes on the power meter. _______________________________________________ pm-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pm-discuss
