Using temperature margin to balance performance with power allocation

Patent metadata

Field	Value
Publication number	US-9710034-B2
Application number	US-201514733164-A
Country	US
Kind code	B2
Filing date	Jun 8, 2015
Priority date	Dec 21, 2012
Publication date	Jul 18, 2017
Grant date	Jul 18, 2017

Abstract

A method and apparatus using temperature margin to balance performance with power allocation. Nominal, middle and high power levels are determined for compute elements. A set of temperature thresholds are determined that drive the power allocation of the compute elements towards a balanced temperature profile. For a given workload, temperature differentials are determined for each of the compute elements relative the other compute elements, where the temperature differentials correspond to workload utilization of the compute element. If temperature overhead is available, and a compute element is below a temperature threshold, then particular compute elements are allocated power to match or drive toward the balanced temperature profile.

First claim

What is claimed is:

1. A power management system, comprising:
a plurality of compute units, each compute unit having a temperature profile;
a microcontroller configured to use temperature measurements of each of the plurality of compute units relative to temperature measurements of at least another compute unit of the plurality of compute units to infer utilization for each of the plurality of compute units for a given workload; and
the microcontroller further configured to allocate power to at least one of the plurality of compute units on a condition that a temperature measurement for at least another of the plurality of compute units is below a temperature threshold, wherein the temperature threshold is based on a temperature of each compute unit and drives toward a temperature profile for an optimal performance for each compute unit for the given workload, wherein the optimal performance represents that each compute unit is not operating at maximum capability or being throttled when additional power is allocated to a compute unit,
wherein the at least one of the plurality of compute units and the at least another of the plurality of compute units are different types of compute units; and
wherein a selected temperature profile from a plurality of temperature profiles allocates power between the at least one of the plurality of compute units and the at least another of the plurality of compute units.

2. The power management system of claim 1 , wherein a first group of the plurality of compute units comprises a plurality of central processing unit (CPU) cores and a second group of the plurality of compute units comprises a plurality of graphics processing units (GPU) cores.

3. The power management system of claim 2 , wherein a selected temperature profile allocates power between the first group and the second group.

4. The power management system of claim 1 , wherein the plurality of compute units share a cooling system.

5. The power management system of claim 1 , wherein the power allocation is made when the at least one compute unit has a temperature overhead.

6. The power management system of claim 1 , wherein the temperature thresholds correspond to a power level between a nominal power level and a maximum power level.

7. The power management system of claim 1 , wherein the temperature threshold is set to indicate less than full utilization of the at least another compute unit.

8. The power management system of claim 1 , wherein a change in the power allocation results in different operating frequency.

9. The power management system of claim 1 , wherein the temperature profile is a balanced temperature profile for the plurality of compute units.

10. A method for power allocation, comprising:
using temperature measurements of each of a plurality of compute units relative to temperature measurements of other compute units in the plurality of compute units to infer utilization for each of the plurality of compute units for a given workload; and
allocating power to at least one of the plurality of compute units on a condition that a temperature measurement for at least another of the plurality of compute units is below a temperature threshold, wherein the temperature threshold is based on a temperature profile of each compute unit and drives toward a temperature profile for an optimal performance for each compute unit for the given workload, wherein the optimal performance represents that each compute unit is not operating at maximum capability or being throttled when additional power is allocated to a compute unit,
wherein the at least one of the plurality of compute units and the at least another of the plurality of compute units are different types of compute units; and
wherein a selected temperature profile from a plurality of temperature profiles allocates power between the at least one of the plurality of compute units and the at least another of the plurality of compute units.

11. The method of claim 10 , wherein a first group of the plurality of compute units comprises a plurality of central processing unit (CPU) cores and a second group of the plurality of compute units comprises a plurality of graphics processing units (GPU) cores.

12. The method of claim 11 , wherein a selected temperature profile allocates power between the first group and the second group.

13. The method of claim 10 , wherein the compute units share a cooling system.

14. The method of claim 10 , wherein the power allocation is made when the at least one compute unit has a temperature overhead.

15. The method of claim 10 , wherein the temperature thresholds correspond to a power level between a nominal power level and a maximum power level.

16. The method of claim 10 , wherein the temperature threshold is set to indicate less than full utilization of the at least another compute element.

17. The method of claim 10 , wherein a change in the power allocation results in different operating frequency.

18. The method of claim 10 , wherein the temperature profile is a balanced temperature profile for the plurality of compute units.

Assignees

• Advanced Micro Devices Inc

Inventors

• Naffziger Samuel D
• Osborn Michael
• Nussbaum Sebastien

Classifications

• G06F1/26 Primary

  Power supply means, e.g. regulation thereof (for memories G11C) · CPC title

• G06F1/206

  comprising thermal management · CPC title