Adaptive power control loop

Patent metadata

Field	Value
Publication number	US-10649518-B2
Application number	US-201715416955-A
Country	US
Kind code	B2
Filing date	Jan 26, 2017
Priority date	Jan 26, 2017
Publication date	May 12, 2020
Grant date	May 12, 2020

Abstract

A GPU performs dynamic power level management by switching between pre-defined power levels having distinct clock and voltage levels. The dynamic power level management includes identifying a first performance metric associated with processing workloads at the for a consecutive number of measurement cycles. In some embodiments, the consecutive number of measurement cycles includes a current measurement cycle and at least one previous measurement cycle. Based on a determination that the consecutive number of measurement cycles exceeds a minimum hysteresis number, an estimated optimization is determined to be applied to the GPU for a future measurement cycle. A power level setting at the GPU for the future measurement cycle is adjusted based on the estimated optimization. By considering performance metrics including, for example, different processing workloads and hardware configurations, the GPU is able to dynamically adapt its power settings to the particular workload that it is currently processing.

First claim

What is claimed is:

1. A method comprising:
identifying a first performance metric associated with processing a current workload at a graphics processing unit (GPU) has a first relationship to a threshold for a number of consecutive measurement cycles, wherein the number of consecutive measurement cycles includes a current measurement cycle and at least one previous measurement cycle, the first performance metric indicative of GPU performance while processing the current workload;
in response to the number of consecutive measurement cycles exceeding one of an up-hysteresis level and a down-hysteresis level,
determining, based on data representing a hardware configuration of a current computing environment processing the current workload, an estimated optimization to be applied to the GPU for a future measurement cycle; and
adjusting, based on the estimated optimization, a power level setting at the GPU by adjusting one of the up-hysteresis level and the down-hysteresis level for the future measurement cycle.

2. The method of claim 1 , wherein adjusting the power level setting comprises changing at least one of a core clock frequency and a memory clock frequency at the GPU.

3. The method of claim 1 , wherein identifying the first performance metric comprises identifying at least one of an average utilization percentage of the GPU, an average operating temperature of the GPU, and an average accumulated power consumed during the current measurement cycle.

4. The method of claim 1 , wherein identifying the first performance metric comprises identifying the first performance metric based on a current processing workload of the GPU for the current measurement cycle and based on an expected future processing workload of the GPU for the future measurement cycle.

5. The method of claim 1 , wherein adjusting the power level setting comprises adjusting the up-hysteresis level.

6. The method of claim 1 , wherein adjusting the power level setting comprises adjusting the down-hysteresis level.

7. The method of claim 1 , wherein adjusting the power level setting at the GPU comprises adjusting the power level setting for the future measurement cycle by a magnitude of a current power level at the GPU based on a difference between the identified first performance metric and a performance metric threshold.

8. A non-transitory computer readable medium embodying a set of executable instructions, the set of executable instructions to manipulate a processor to:
identify a first performance metric associated with processing a current workload at a graphics processing unit (GPU) has a first relationship to a threshold for a number of consecutive measurement cycles, wherein the number of consecutive measurement cycles includes a current measurement cycle and at least one previous measurement cycle, the first performance metric indicative of GPU performance while processing the current workload;
in response to the number of consecutive measurement cycles exceeding one of an up-hysteresis level and a down-hysteresis level,
determine an estimated optimization to be applied to the GPU for a future measurement cycle; and
adjust a power level setting at the GPU for the future measurement cycle by adjusting one of the up-hysteresis level and the down-hysteresis level based on the estimated optimization.

9. The non-transitory computer readable medium of claim 8 , wherein adjusting the power level setting comprises changing at least one of a core clock frequency and a memory clock frequency at the GPU.

10. The non-transitory computer readable medium of claim 8 , wherein identifying the first performance metric comprises identifying at least one of an average utilization percentage of the GPU, an average operating temperature of the GPU, and an average accumulated power consumed during the current measurement cycle.

11. The non-transitory computer readable medium of claim 8 , wherein adjusting the power level comprises adjusting the up-hysteresis level.

12. The non-transitory computer readable medium of claim 8 , wherein adjusting the power level comprises adjusting the down-hysteresis level.

13. A device, comprising:
a graphics processing unit (GPU) comprising:
a graphics pipeline including a plurality of compute units (CUs);
a performance module to:
identify a first performance metric associated with processing a current workload at the GPU has a first relationship with a threshold for a number of consecutive measurement cycles, wherein the consecutive number of consecutive measurement cycles includes a current measurement cycle and at least one previous measurement cycle, the first performance metric indicative of GPU performance while processing the current workload;
in response to the number of consecutive measurement cycles exceeding one of an up-hysteresis level and a down-hysteresis level, determine, based on data representing a hardware configuration of the device, an estimated optimization to be applied to the GPU for a future measurement cycle; and

a power and clock controller module to adjust a power level setting at the GPU based on the estimated optimization by adjusting one of the up-hysteresis level and the down-hysteresis level.

14. The device of claim 13 , wherein the estimated optimization is selected from a set of pre-defined power level settings including a plurality of performance optimized setting and a plurality of power savings optimized settings.

15. The device of claim 14 , wherein the power and clock controller module is further configured to adjust at least one of a core clock frequency and a memory clock frequency at the GPU based on the estimated optimization.

16. The device of claim 13 , wherein the performance module is further configured to identify at least one of an average utilization percentage of the GPU, an average operating temperature of the GPU, and an average accumulated power consumed during the current measurement cycle.

17. The device of claim 13 , wherein the device further comprises a GPU firmware configured to calculate the first performance metric based on sensor data captured at the performance module.

Assignees

• Ati Technologies Ulc

Inventors

• Kwon Soon Kyu
• Huang Jun
• Pezeshgi Shahriar
• Duenas Alexander Sabino

Classifications

• G06F1/3218

  of display devices · CPC title

• G06F1/3234 Primary

  Power saving characterised by the action undertaken · CPC title

• G06F1/324 Primary

  by lowering clock frequency · CPC title

• G06T1/20

  Processor architectures; Processor configuration, e.g. pipelining · CPC title

• G06F1/3206

  Monitoring of events, devices or parameters that trigger a change in power modality · CPC title