Granted Patent B2

US 12,235,711 · App. 18/250,417 · Granted Feb 25, 2025

Inventors: Junxun Wu (Jiangsu, CN); Guanmin Huang (Jiangsu, CN)

Assignee: INSPUR SUZHOU INTELLIGENT TECHNOLOGY CO., LTD.

G06F1/3287

Abstract

Claims (66)

1. A power supply control method for a server, comprising:

dividing a utilization rate of a system main power supply into different levels in advance, and setting a Graphics Processing Unit (GPU) power control policy corresponding to a respective one of the different levels of the utilization rate of the system main power supply, wherein the level of the utilization rate of the system main power supply increases with the increase in the utilization rate of the system main power supply, and a suppression degree, on a computing capability of GPUs in a system, of the set GPU power control policy increases with the increase in the level of the utilization rate of the system main power supply;

acquiring an actual utilization rate of the system main power supply, and determining, according to a result of the level division on the utilization rate of the system main power supply, a target utilization rate level corresponding to the actual utilization rate; and

performing power supply control on the GPUs in the system according to the GPU power control policy corresponding to the target utilization rate level.

2. The power supply control method for the server according to claim 1 ,

wherein performing the level division on the utilization rate of the system main power supply in advance, and setting the GPU power control policy corresponding to the respective one of the different levels of the utilization rate of the system main power supply comprises:

dividing the utilization rate of the system main power supply into three levels in advance, to obtain a low-level utilization rate, a medium-level utilization rate and a high-level utilization rate;

setting, corresponding to the low-level utilization rate, a first GPU power control policy used for adjusting load balance distribution of the GPUs in the system;

setting, corresponding to the medium-level utilization rate, a second GPU power control policy used for selecting a target trigger GPU from the GPUs in the system according to a preset GPU trigger selection policy, and triggering a power brake signal for the target trigger GPU; and

setting, corresponding to the high-level utilization rate, a third GPU power control policy used for selecting a target turn-off GPU from the GPUs in the system according to a preset GPU turn-off selection policy and turning off power supply of the target turn-off GPU.

3. The power supply control method for the server according to claim 2 ,

wherein adjusting the load balance distribution of the GPUs in the system comprises:

pre-configuring, for a respective GPU of the GPUs in the system, an integrated chip for acquiring power supply parameters of the respective GPU, and acquiring a current parameter of the respective GPU by using the integrated chip;

determining whether the current parameter of a target GPU is greater than a preset current threshold; when the current parameter of the target GPU is greater than the preset current threshold, determining that the target GPU is in a high workload state; and when the current parameter of the target GPU is not greater than the preset current threshold, determining that the target GPU is in a low workload state, wherein the target GPU is any GPU of the GPUs in the system; and

distributing, in a balanced manner, a part of computing workloads of GPUs in the high workload state to GPUs in the low workload state, so that the GPUs originally in the high workload state transit to the low workload state.

4. The power supply control method for the server according to claim 3 , further comprising:

respectively comparing a respective one of the power supply parameters of the target GPU acquired by using the integrated chip with a preset parameter security threshold corresponding to the respective power supply parameter; and when any power supply parameter of the target GPU is greater than the preset parameter security threshold corresponding to the power supply parameter, directly turning off the power supply of the target GPU by using the integrated chip.

5. The power supply control method for the server according to claim 2 , wherein selecting the target trigger GPU from the GPUs in the system according to the preset GPU trigger selection policy, and triggering the power brake signal for the target trigger GPU comprises:

pre-configuring, for a respective GPU of the GPUs in the system, an integrated chip for acquiring power supply parameters of the respective GPU, and acquiring a current parameter of the respective GPU by using the integrated chip;

determining whether the current parameter of a target GPU is greater than a preset current threshold, and when the current parameter of the target GPU is greater than the preset current threshold, determining that the target GPU is in a high workload state, wherein the target GPU is any GPU of the GPUs in the system; and

sequentially triggering the power brake signal for a respective GPU of GPUs in the high workload state, until the actual utilization rate of the system main power supply decreases to the low-level utilization rate.

6. The power supply control method for the server according to claim 5 , further comprising:

respectively comparing a respective one of the power supply parameters of the target GPU acquired by using the integrated chip with a preset parameter security threshold corresponding to the respective power supply parameter; and when any power supply parameter of the target GPU is greater than the preset parameter security threshold corresponding to the power supply parameter, directly turning off the power supply of the target GPU by using the integrated chip.

7. The power supply control method for the server according to claim 2 , wherein selecting the target turn-off GPU from the GPUs in the system according to the preset GPU turn-off selection policy and turning off the power supply of the target turn-off GPU comprises:

pre-configuring, for a respective GPU of the GPUs in the system, an integrated chip for acquiring power supply parameters of the respective GPU, and acquiring a current parameter of the respective GPU by using the integrated chip;

determining whether the current parameter of a target GPU is greater than a preset current threshold; when the current parameter of the target GPU is greater than the preset current threshold, determining that the target GPU is in a high workload state; and when the current parameter of the target GPU is not greater than the preset current threshold, determining that the target GPU is in a low workload state, wherein the target GPU is any GPU of the GPUs in the system; and

first turning off power supply of GPUs in the low workload state, and then turning off power supply of GPUs in the high workload state, until the actual utilization rate of the system main power supply decreases to the medium-level utilization rate.

8. The power supply control method for the server according to claim 7 , further comprising:

respectively comparing a respective one of the power supply parameters of the target GPU acquired by using the integrated chip with a preset parameter security threshold corresponding to the respective power supply parameter; and when any power supply parameter of the target GPU is greater than the preset parameter security threshold corresponding to the power supply parameter, directly turning off the power supply of the target GPU by using the integrated chip.

9. The power supply control method for the server according to claim 7 , wherein first turning off power supply of GPUs in the low workload state, and then turning off power supply of GPUs in the high workload state, until the actual utilization rate of the system main power supply decreases to the medium-level utilization rate comprises:

first sequentially turning off the power supply of the GPUs in the low workload state, and then sequentially turning off the power supply of GPUs in the high workload state, until the actual utilization rate of the system main power supply decreases to the medium-level utilization rate.

10. The power supply control method for the server according to claim 1 , wherein dividing the utilization rate of the system main power supply into different levels in advance comprises:

dividing the utilization rate, which is equal to or above a preset threshold, of the system main power supply into different levels in advance.

11. The power supply control method for the server according to claim 1 , wherein determining, according to the result of the level division on the utilization rate of the system main power supply, the target utilization rate level corresponding to the actual utilization rate comprises:

determining, according to a correlation between the level of the utilization rate of the system main power supply and the GPU power control policy, the target utilization rate level corresponding to the actual utilization rate.

12. A power supply control system for a server, comprising a Power Supply Unit (PSU) and a power supply control device, wherein the power supply control device comprises: a processor connected between the PSU and Graphics Processing Units (GPUs) and configured to implement following operations when a computer program stored in a memory is executed:

dividing a utilization rate of a system main power supply into different levels in advance, and setting a (GPU) power control policy corresponding to a respective one of the different levels of the utilization rate of the system main power supply, wherein the level of the utilization rate of the system main power supply increases with the increase in the utilization rate of the system main power supply, and a suppression degree, on a computing capability of GPUs in a system, of the set GPU power control policy increases with the increase in the level of the utilization rate of the system main power supply;

acquiring an actual utilization rate of the system main power supply, and determining, according to a result of the level division on the utilization rate of the system main power supply, a target utilization rate level corresponding to the actual utilization rate; and

performing power supply control on the GPUs in the system according to the GPU power control policy corresponding to the target utilization rate level.

13. The power supply control system for the server according to claim 12 , wherein the power supply control device further comprises integrated chips respectively connected with the GPUs, and the integrated chips are used for acquiring power supply parameters of the respective GPUs.

14. The power supply control system for the server according to claim 13 , wherein the integrated chips are Hotswap Integrated Circuits (ICs).

15. A power supply control device for a server, comprising:

a memory, configured to store a computer program; and

a processor, provided between a system main power supply and Graphics Processing Units (GPUs) in a system, and configured to implement following operations when executing the computer program:

dividing a utilization rate of a system main power supply into different levels in advance, and setting a Graphics Processing Unit (GPU) power control policy corresponding to a respective one of the different levels of the utilization rate of the system main power supply, wherein the level of the utilization rate of the system main power supply increases with the increase in the utilization rate of the system main power supply, and a suppression degree, on a computing capability of GPUs in a system, of the set GPU power control policy increases with the increase in the level of the utilization rate of the system main power supply;

acquiring an actual utilization rate of the system main power supply, and determining, according to a result of the level division on the utilization rate of the system main power supply, a target utilization rate level corresponding to the actual utilization rate; and

performing power supply control on the GPUs in the system according to the GPU power control policy corresponding to the target utilization rate level.

16. The power supply control device for the server according to claim 15 , wherein performing the level division on the utilization rate of the system main power supply in advance, and setting the GPU power control policy corresponding to the respective one of the different levels of the utilization rate of the system main power supply comprises:

dividing the utilization rate of the system main power supply into three levels in advance, to obtain a low-level utilization rate, a medium-level utilization rate and a high-level utilization rate;

setting, corresponding to the low-level utilization rate, a first GPU power control policy used for adjusting load balance distribution of the GPUs in the system;

setting, corresponding to the medium-level utilization rate, a second GPU power control policy used for selecting a target trigger GPU from the GPUs in the system according to a preset GPU trigger selection policy, and triggering a power brake signal for the target trigger GPU; and

setting, corresponding to the high-level utilization rate, a third GPU power control policy used for selecting a target turn-off GPU from the GPUs in the system according to a preset GPU turn-off selection policy and turning off power supply of the target turn-off GPU.

17. The power supply control device for the server according to claim 16 , wherein adjusting the load balance distribution of the GPUs in the system comprises:

pre-configuring, for a respective GPU of the GPUs in the system, an integrated chip for acquiring power supply parameters of the respective GPU, and acquiring a current parameter of the respective GPU by using the integrated chip;

determining whether the current parameter of a target GPU is greater than a preset current threshold; when the current parameter of the target GPU is greater than the preset current threshold, determining that the target GPU is in a high workload state; and when the current parameter of the target GPU is not greater than the preset current threshold, determining that the target GPU is in a low workload state, wherein the target GPU is any GPU of the GPUs in the system; and

distributing, in a balanced manner, a part of computing workloads of GPUs in the high workload state to GPUs in the low workload state, so that the GPUs originally in the high workload state transit to the low workload state.

18. The power supply control device for the server according to claim 17 , wherein the processor is further configured to implement following operation when executing the computer program:

respectively comparing a respective one of the power supply parameters of the target GPU acquired by using the integrated chip with a preset parameter security threshold corresponding to the respective power supply parameter; and when any power supply parameter of the target GPU is greater than the preset parameter security threshold corresponding to the power supply parameter, directly turning off the power supply of the target GPU by using the integrated chip.

19. The power supply control device for the server according to claim 16 , wherein selecting the target trigger GPU from the GPUs in the system according to the preset GPU trigger selection policy, and triggering the power brake signal for the target trigger GPU comprises:

pre-configuring, for a respective GPU of the GPUs in the system, an integrated chip for acquiring power supply parameters of the respective GPU, and acquiring a current parameter of the respective GPU by using the integrated chip;

determining whether the current parameter of a target GPU is greater than a preset current threshold, and when the current parameter of the target GPU is greater than the preset current threshold, determining that the target GPU is in a high workload state, wherein the target GPU is any GPU of the GPUs in the system; and

sequentially triggering the power brake signal for a respective GPU of GPUs in the high workload state, until the actual utilization rate of the system main power supply decreases to the low-level utilization rate.

20. The power supply control device for the server according to claim 16 , wherein selecting the target turn-off GPU from the GPUs in the system according to the preset GPU turn-off selection policy and turning off the power supply of the target turn-off GPU comprises:

pre-configuring, for a respective GPU of the GPUs in the system, an integrated chip for acquiring power supply parameters of the respective GPU, and acquiring a current parameter of the respective GPU by using the integrated chip;

determining whether the current parameter of a target GPU is greater than a preset current threshold; when the current parameter of the target GPU is greater than the preset current threshold, determining that the target GPU is in a high workload state; and when the current parameter of the target GPU is not greater than the preset current threshold, determining that the target GPU is in a low workload state, wherein the target GPU is any GPU of the GPUs in the system; and

first turning off power supply of GPUs in the low workload state, and then turning off power supply of GPUs in the high workload state, until the actual utilization rate of the system main power supply decreases to the medium-level utilization rate.

Assignments (2)

Priority Claims (1)

Continuity (1)

References Cited (17)