Intel® Xeon Phi™ coprocessor Power Management Configuration: Why should I worry about configuring anything?

Power setting change

Reason

Turbo off

1. Your application is particularly sensitive to OS jitter
2. Your cluster cannot exceed a certain fixed power ceiling
3. Your application needs a constant frequency (meaning it is sensitive to variations in the frequency of the processor)

Turbo on

1. You want your application to get every performance boast that is possible
2. Your application has (relatively) frequent periods of “rest” such that turbo has the potential of providing a significant performance boost.

PC6 enable

1. Your application has no special thermal or latency requirements
2. Your application has long periods of inactivity
3. Your cluster has a low thermal ceiling and relatively long periods of coprocessor inactivity

PC6 disable

1. Your application cannot tolerate the high latency inherent in reactivating an application between periods of low activity
2. Your application cannot ignore coprocessor directed PCI events

PC3 enable / disable

Examples similar to that of PC6 enable/disable with the exception of PCI events. Latency dramatically increases once you drop into PC3 since the part of the coprocessor power management module that is on the host needs to initiate an exit from PC3 back to C0.

P1 enable / disable

1. The cooling capability of your host/coprocessor system needs to be “smoothed” to avoid thermal spikes (enable)
2. Your application has no special thermal requirements (enable)
3. Your application needs a constant frequency, meaning it is sensitive to variations in the frequency of the processor (disable)

CO6 enable / disable

1. Your application is particularly sensitive even to short latencies (disable)
2. Your application has no special thermal or latency requirements (enable)