Methods, apparatus, and systems to dynamically schedule workloads among compute resources based on temperature

Patent metadata

Field	Value
Publication number	US-11972303-B2
Application number	US-202016914177-A
Country	US
Kind code	B2
Filing date	Jun 26, 2020
Priority date	Jun 26, 2020
Publication date	Apr 30, 2024
Grant date	Apr 30, 2024

Abstract

Methods, apparatus, and systems to dynamically schedule a workload to among compute blocks based on temperature are disclosed. An apparatus to schedule a workload to at least one of a plurality of compute blocks based on temperature includes a prediction engine to determine (i) a first predicted temperature of a first compute block of the plurality of compute blocks and (ii) a second predicted temperature of a second compute block of the plurality of compute blocks. The apparatus also includes a selector to select between the first compute block and the second compute block for assignment of the workload. The selection is based on which of the first and second predicted temperatures is lower. The apparatus further includes a workload scheduler to assign the workload to the selected one of the first or second compute blocks.

First claim

What is claimed is:

1. An apparatus to schedule a workload, the apparatus comprising:
interface circuitry;
instructions; and
processor circuitry to operate based on the instructions to:
generate a heat pattern associated with a plurality of compute blocks of a compute device; and
assign the workload to at least one of a first compute block or a second compute block of the plurality of compute blocks, the workload to also utilize a third compute block of the plurality of compute blocks, the assignment based on the heat pattern, a first distance between the first compute block and the third compute block, and a second distance between the second compute block and the third compute block.

2. The apparatus of claim 1 , wherein the assigned at least one of the first compute block or the second compute block is to experience less power leakage when executing the workload than the one of the first and second compute blocks not selected.

3. The apparatus of claim 1 , wherein the processor circuitry is to predict a first predicted temperature of the first compute block and a second predicted temperature of the second compute block based on (i) the heating pattern, (ii) a first predicted change to the heating pattern if the workload is assigned to the first compute block and (iii) a second predicted change to the heating pattern if the workload is assigned to the second compute block.

4. The apparatus of claim 1 , wherein the processor circuitry is to:
generate a first predicted temperature of the first compute block based on a first operating power utilization parameter of the first compute block; and
generate a second predicted temperature of the second compute block based on a second operation power utilization parameter of the second compute block.

5. The apparatus of claim 1 , wherein the processor circuitry is to:
model a thermal flow on a silicon die associated with the plurality of compute blocks; and
determine a first predicted temperature of the first compute block and a second predicted temperature of the second compute block based on the thermal flow.

6. The apparatus of claim 5 , wherein the processor circuitry is to determine a change in temperature between a third temperature of the first compute block and the first predicted temperature of the first compute block, the third temperature determined before the workload is assigned to any of the plurality of compute blocks.

7. The apparatus of claim 1 , wherein the heat pattern is to indicate respective temperatures of corresponding regions of a silicon die, the plurality of compute blocks on the silicon die.

8. The apparatus of claim 1 , wherein the processor circuitry is to:
predict a first change of a first temperature of the first compute block based on assignment of the workload to the first compute block; and
predict a second change of a second temperature of the second compute block based on assignment of the workload to the second compute block.

9. The apparatus of claim 1 , wherein the processor circuitry is to determine a first predicted temperature of the first compute block and a second predicted temperature of the second compute block based on the heating pattern.

10. The apparatus of claim 1 , wherein the processor circuitry is to:
calculate a first operating power of the first compute block and a second operating power of the second compute block; and
determine a first change in temperature of the first compute block and a second change in temperature of the second compute block based on the first and second operating powers, respectively.

11. The apparatus of claim 1 , wherein the processor circuitry is to rank the plurality of compute blocks based on respective power efficiencies of corresponding ones of the plurality of compute blocks.

12. One or more non-transitory computer readable medium comprising instructions to cause one or more processors to at least:
perform an initial calibration of a compact transient thermal model using a system computational fluid dynamics model;
update the compact transient thermal model based on changes in utilization of a plurality of compute blocks, the plurality of compute blocks on a silicon die;
generate a first heat pattern of the silicon die based on the compact transient thermal model;
generate a second heat pattern of the silicon die and a third heat pattern of the silicon die based on the compact transient thermal model and the first heat pattern, the second heat pattern based on a first assumption that a workload is assigned to a first compute block of the plurality of compute blocks and the third heat pattern based on a second assumption that the workload is assigned to a second compute block of the plurality of compute blocks; and
assign a workload to at least one of the first compute block or the second compute block based on the second heat pattern and the third heat pattern.

13. The one or more non-transitory computer readable medium of claim 12 , wherein the first, second, and third heat patterns reflect temperatures at locations on the silicon die due to workloads performed by the plurality of compute blocks, and the instructions are to cause the one or more processors to:
based on a fourth heat pattern, determine a temperature at a first location of the silicon die at which a third compute block is disposed and at a second location of the silicon die that is adjacent to the third compute block; and
determine whether to assign a second workload to the third compute block based on the temperature at the first location and at the second location.

14. The one or more non-transitory computer readable medium of claim 12 , wherein the instructions are to cause the one or more processors to:
generate a plurality of heat patterns including the second and third heat patterns, the plurality of heat patterns generated based on the compact transient thermal model;
generate a new heat pattern to reflect a new workload assignment when a new workload is assigned to any of the compute blocks; and
rank the plurality of compute blocks based on the new heat pattern, respective ones of the plurality of compute blocks ranked based on respective real-time power efficiencies of corresponding ones of the plurality of compute blocks.

15. The one or more non-transitory computer readable medium of claim 12 , wherein the workload is a first workload, and the instructions are to cause the one or more processors to:
recommend that a second workload be assigned to a third compute block based on a fourth heating pattern; and
override a policy to cause the second workload to be assigned based on the recommendation.

16. The one or more non-transitory computer readable medium of claim 12 , wherein the first heat pattern reflects a pattern of respective temperatures on respective regions of the silicon die caused by existing workloads executed by one or more of the compute blocks.

17. The one or more non-transitory computer readable medium of claim 12 , wherein the instructions are to cause the one or more processors to compare a map of respective locations of corresponding ones of the plurality of compute blocks on the silicon die to the first heat pattern to correlate respective temperatures of the first heat pattern to corresponding ones of the compute blocks of the silicon die.

18. A method to schedule a workload, the method comprising:
determining, by executing an instruction with at least one processor, a first predicted temperature of a first compute block of a plurality of compute blocks of a compute device based on a sensed board tempe

Assignees

• Intel Corp

Inventors

• Ruiz Carin
• Qiu Bo
• Mishra Columbia
• Chattopadhyay Arijit
• Nge Chee Lim
• Potluri Srikanth
• Zhu Jianfang
• Kirubakaran Deepak Samuel
• Rallabandi Akhilesh
• Gallina Mark
• Thomas Renji
• Hermerding Ii James

Classifications

• G06F9/5094 Primary

  where the allocation takes into account power or heat criteria (power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206) · CPC title

• G06F9/4881

  Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues · CPC title

• G06F9/505

  considering the load · CPC title

• G06F11/3058 Primary

  Monitoring arrangements for monitoring environmental properties or parameters of the computing system or of the computing system component, e.g. monitoring of power, currents, temperature, humidity, position, vibrations (thermal management in cooling arrangements of a computing system G06F1/206) · CPC title

• G06F2209/508

  Monitor · CPC title