Integrated circuit performance adaptation using workload predictions
Performance adaptation for an integrated circuit includes receiving, by a workload prediction system of a hardware processor, telemetry data for one or more systems of the hardware processor. A workload prediction is determined by processing the telemetry data through a workload prediction model executed by a workload prediction controller of the workload prediction system. A profile is selected, from a plurality of profiles, that matches the workload prediction. The selected profile specifies one or more operating parameters for the hardware processor. The selected profile is provided to a power management controller of the hardware processor for controlling an operational characteristic of the one or more systems.
1. A method, comprising:
receiving, by a workload prediction system of a hardware processor embodied as an integrated circuit, telemetry data for one or more systems of the hardware processor;
determining a workload prediction by processing the telemetry data through a workload prediction model executed by a workload prediction controller of the workload prediction system, wherein the workload prediction controller is embodied as a hardware controller embedded in the integrated circuit and dedicated to performing workload prediction;
selecting a profile, from a plurality of profiles, that matches the workload prediction, wherein the selected profile specifies one or more operating parameters for the hardware processor; and
providing the selected profile to a power management controller of the hardware processor for controlling an operational characteristic of the one or more systems.
2. The method of claim 1 , wherein the workload prediction specifies a candidate application likely executing in the hardware processor as of a time corresponding to the telemetry data.
3. The method of claim 2 , wherein the candidate application is a particular application selected from a class of application that includes a plurality of different candidate applications.
4. The method of claim 2 , wherein the selected profile is one of a plurality of different profiles corresponding to the candidate application, wherein each profile of the plurality of profiles of the application corresponds to a different operating state of the candidate application.
5. The method of claim 1 , further comprising:
adjusting, by the power management controller, the operating characteristic of the one or more systems in response to, and based on, the selected profile.
6. The method of claim 1 , wherein the receiving the telemetry data, the determining the workload prediction, the matching the workload prediction with the profile, and the providing the selected profile to the power management controller are performed in real-time.
7. The method of claim 1 , wherein the receiving the telemetry data, the determining the workload prediction, the matching the workload prediction with the profile, and the providing the selected profile to the power management controller are performed continuously.
8. The method of claim 1 , wherein the workload prediction model is a machine learning model.
9. The method of claim 8 , wherein the machine learning model is trained to select a candidate application from a plurality of candidate applications based on the telemetry data.
10. The method of claim 9 , wherein the machine learning model is further trained to select an operating state of the candidate application from a plurality of operating states for the candidate application, and wherein the selected profile further depends on the selected operating state of the candidate application.
11. A hardware processor, comprising:
a system management unit; and
one or more systems coupled to the system management unit;
wherein the hardware processor is embodied as an integrated circuit; and
wherein the system management unit includes:
a workload prediction controller embodied as a hardware controller embedded in the integrated circuit and dedicated to performing workload prediction; and
a power management controller;
wherein the workload prediction controller is configured to execute operations including:
receiving telemetry data for the one or more systems;
determining a workload prediction by processing the telemetry data through a workload prediction model executed by the workload prediction controller;
selecting a profile, from a plurality of profiles, that matches the workload prediction, wherein the selected profile specifies one or more operating parameters for the hardware processor; and
providing the selected profile to the power management controller of the hardware processor for controlling an operational characteristic of the one or more systems.
12. The hardware processor of claim 11 , wherein the workload prediction specifies a candidate application likely executing in the hardware processor as of a time corresponding to the telemetry data.
13. The hardware processor of claim 12 , wherein the candidate application is a particular application selected from a class of application that includes a plurality of different candidate applications.
14. The hardware processor of claim 12 , wherein the selected profile is one of a plurality of different profiles corresponding to the candidate application, wherein each profile of the plurality of profiles of the candidate application corresponds to a different operating state of the candidate application.
15. The hardware processor of claim 11 , wherein the power management controller is configured to adjust the operating characteristic of the one or more systems in response to, and based on, the selected profile.
16. The hardware processor of claim 11 , wherein the receiving the telemetry data, the determining the workload prediction, the matching the workload prediction with the profile, and the providing the selected profile to the power management controller, as performed by the workload prediction controller, are performed in real-time.
17. The hardware processor of claim 11 , wherein the receiving the telemetry data, the determining the workload prediction, the matching the workload prediction with the profile, and the providing the selected profile to the power management controller, as performed by the workload prediction controller, are performed continuously.
18. The hardware processor of claim 11 , wherein the workload prediction model is a machine learning model.
19. The hardware processor of claim 18 , wherein the machine learning model is trained to select a candidate application from a plurality of candidate applications based on the telemetry data.
20. The hardware processor of claim 19 , wherein the machine learning model is further trained to select an operating state of the candidate application from a plurality of operating states for the candidate application, and wherein the selected profile further depends on the selected operating state of the candidate application.