摘要:
Methods, apparatus, and computer program products are described for dynamic frequency and voltage scaling for a computer processor, including identifying during manufacture a nominal operating point of frequency and voltage for a computer processor, the nominal operating point including a nominal operating voltage identified for a design nominal operating frequency; determining, in dependence upon the nominal operating point, an operating range of frequency and voltage over which the computer processor is to function; and storing, in non-volatile storage on the computer processor during manufacture, information specifying the nominal operating point and the operating range.
摘要:
A system for identifying a subset of sensors to sample to reduce the frequency of sensor access. The system determines rise times and records values for the sensors in the system. A time criticality of the sensors is determined based on the rise times. The system processes the sensors by first creating sensor subsets based on one or more constraints on the sensors. The system monitors the values of the sensors in a sensor subset and flags a sensor when it makes a determination that, prior to a next scheduled sampling of the sensor subset, the value of a sensor in the monitored sensor subset will exceed a threshold constraint. The system moves those flagged sensors to a second sensor subset which complies with the sensor's constraints.
摘要:
A weighted event counting system and method for processor performance measurements provides low latency and low error performance measurement capability. A weighted performance counter accumulates a performance count according to a plurality of event signals provided from functional units in the processor. Differing weights are applied to the event signals in according to the correlation between each event with processor performance. The weights may be provided from programmable registers, so that the weights can be adjusted under program control. The event signals may be combined to reduce the bit-width of the set of event signal, with mutually-exclusive events merged in single fields of the combinatorial result and events having the same weights merged according to a sub-total. The weights are applied to the combinatorial result and used to update a performance count. The performance count can then be used by power management software or hardware to make adjustments in operating parameters of the processor.
摘要:
A method and system for improving processing performance by using activity factor headroom provides improved performance while meeting power management constraints in a processing system. The method and system estimate the power consumption of the system from a model that relates measured activities at a present operating point to power consumption for any available operating point of one or more processors in the system. The method then chooses the operating point(s) with the highest performance among the available operating points that will still meet budgetary constraints or specific thresholds of power consumption. The budgetary constraints or specific thresholds may be dynamically adjusted, and the method will update the operating point(s) to maintain safe operation and maximize performance. The method provides the best performance for the executing workload while ensuring safe operation.
摘要:
During manufacture, an operating range for dynamic voltage and frequency scaling can be established. A nominal operating point is identified based on a design nominal operating frequency for a computer processor. The nominal operating point comprises a nominal operating voltage identified for the design nominal operating frequency. In dependence upon the nominal operating point, an operating range of frequency and voltage over which the computer processor is to function is determined. Information specifying the nominal operating point and the operating range is stored in non-volatile storage associated with the computer processor.
摘要:
A system for identifying a subset of sensors to sample to reduce the frequency of sensor access. The system determines rise times and records values for the sensors in the system. A time criticality of the sensors is determined based on the rise times. The system processes the sensors by first creating sensor subsets based on one or more constraints on the sensors. The system monitors the values of the sensors in a sensor subset and flags a sensor when it makes a determination that, prior to a next scheduled sampling of the sensor subset, the value of a sensor in the monitored sensor subset will exceed a threshold constraint. The system moves those flagged sensors to a second sensor subset which complies with the sensor's constraints.
摘要:
A benchmark tester retrieves a voltage margin that corresponds to a device that a system includes. The voltage margin indicates an additional amount of voltage to apply to a nominal voltage that, when added, results in the device operating at a power limit while executing a worst-case power workload. Next, the benchmark tester (or thermal power management device) sets an input voltage for the device to a value equal to the sum of the voltage margin and the nominal voltage. The benchmark tester then dynamically benchmark tests the system, which includes adjusting the device's frequency and input voltage while ensuring that the device does not exceed the device's power limit. In turn, the benchmark tester records a guaranteed minimum performance boost for the system based upon a result of the benchmark testing.
摘要:
A system is provided for unified management of power, performance, and thermals in computer systems. This system incorporates elements to effectively address all aspects of managing computing systems in an integrated manner, instead of independently. The system employs an infrastructure for real-time measurements feedback, an infrastructure for regulating system activity, component operating levels, and environmental control, a dedicated control structure for guaranteed response/preemptive action, and interaction and integration components. The system provides interfaces for user-level interaction. The system also employs methods to address power/thermal concerns at multiple timescales. In addition, the system improves efficiency by adopting an integrated approach, rather than treating different aspects of the power/thermal problem as individual issues to be addressed in a piecemeal fashion.
摘要:
A data processing system includes a power supply, a plurality of processors wherein each processor is separately powerable by the power supply under operating system control. The operating system determines periodically a measure of system utilization and controls the switches to alter the number of active (powered) processors where the number of active processors reflects the measured system utilization and a set of utilization threshold values. System utilization may be based on the number of active tasks. The utilization thresholds preferably include a maximum threshold and a minimum threshold. A measured utilization exceeding the maximum threshold causes an increase in the number of active processors while utilization less than the minimum threshold causes a decrease in the number of active processors. The utilization thresholds may be determined from threshold factors that reflect time and date information, quality of service information, or a weighted average of historical utilization values.
摘要:
A power management system schedules the voltage and frequency of processors in a data processing system based on two criteria. The first criterion is a prediction of the performance that the work currently running on the processor will experience at the different frequencies that are available. The second criterion is a system-wide constraint on the total power budget allocated to processors. Based on these criteria, low-level code sets the frequency and voltage of the processors in the system to match what the operating system is currently running on them.