US 12,591,286 B2Granted

Controlling execution of artificial intelligence workloads based on predicted power consumption

AssigneeDell Products L.P.

InventorsColin Stewart Byrne, Ian Roche

Life of this patent

Estimating the term…

Loading patent…

Abstract

An apparatus comprises at least one processing device configured to determine, for a first set of artificial intelligence (AI) workloads, a first set of metrics characterizing AI models utilized by the first set of AI workloads and a second set of metrics characterizing power consumption associated with execution of the first set of AI workloads. The at least one processing device is also configured to generate, based on the first and second sets of metrics, a predictive model configured for generating power consumption predictions for a second set of AI workloads, to utilize the predictive model to generate, for a given one of the second set of AI workloads, predicted power consumption associated with execution of the given AI workload on one or more information technology (IT) assets, and to control execution of the given AI workload on a given IT asset based on the predicted power consumption.

Claims20

Dependent claims

An apparatus comprising:

at least one processing device comprising a processor coupled to a memory;
the at least one processing device being configured:
to determine, for a first set of one or more artificial intelligence workloads, a first set of metrics characterizing one or more artificial intelligence models utilized by the first set of one or more artificial intelligence workloads and a second set of metrics characterizing power consumption associated with execution of the first set of one or more artificial intelligence workloads;
to generate, based at least in part on the first set of metrics and the second set of metrics, a predictive model configured for generating predictions of power consumption associated with execution of a second set of one or more artificial intelligence workloads;
to utilize the predictive model to generate, for a given one of the second set of one or more artificial intelligence workloads, at least one predicted power consumption associated with execution of the given one of the second set of one or more artificial intelligence workloads on one or more information technology assets, the given one of the second set of one or more artificial intelligence workloads comprising training a given artificial intelligence model, the at least one predicted power consumption associated with the execution of the given one of the second set of one or more artificial intelligence workloads on the one or more information technology assets comprising a per training epoch power consumption prediction for each training epoch of training the given artificial intelligence model;
to determine a training process for the training the given artificial intelligence model based at least in part on the per training epoch power consumption prediction for each training epoch of training the given artificial intelligence model;
and
to control execution of the given one of the second set of one or more artificial intelligence workloads on at least a given one of the one or more information technology assets based at least in part on the at least one predicted power consumption and the determined training process for the training of the given artificial intelligence model.

5 more elements

A computer program product comprising a non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device:

to determine, for a first set of one or more artificial intelligence workloads, a first set of metrics characterizing one or more artificial intelligence models utilized by the first set of one or more artificial intelligence workloads and a second set of metrics characterizing power consumption associated with execution of the first set of one or more artificial intelligence workloads;
to generate, based at least in part on the first set of metrics and the second set of metrics, a predictive model configured for generating predictions of power consumption associated with execution of a second set of one or more artificial intelligence workloads;
to utilize the predictive model to generate, for a given one of the second set of one or more artificial intelligence workloads, at least one predicted power consumption associated with execution of the given one of the second set of one or more artificial intelligence workloads on one or more information technology assets, the given one of the second set of one or more artificial intelligence workloads comprising training a given artificial intelligence model, the at least one predicted power consumption associated with the execution of the given one of the second set of one or more artificial intelligence workloads on the one or more information technology assets comprising a per training epoch power consumption prediction for each training epoch of training the given artificial intelligence model;
to determine a training process for the training the given artificial intelligence model based at least in part on the per training epoch power consumption prediction for each training epoch of training the given artificial intelligence model;
and
to control execution of the given one of the second set of one or more artificial intelligence workloads on at least a given one of the one or more information technology assets based at least in part on the at least one predicted power consumption and the determined training process for the training of the given artificial intelligence model.

3 more elements

A method comprising:

determining, for a first set of one or more artificial intelligence workloads, a first set of metrics characterizing one or more artificial intelligence models utilized by the first set of one or more artificial intelligence workloads and a second set of metrics characterizing power consumption associated with execution of the first set of one or more artificial intelligence workloads;
generating, based at least in part on the first set of metrics and the second set of metrics, a predictive model configured for generating predictions of power consumption associated with execution of a second set of one or more artificial intelligence workloads;
utilizing the predictive model to generate, for a given one of the second set of one or more artificial intelligence workloads, at least one predicted power consumption associated with execution of the given one of the second set of one or more artificial intelligence workloads on one or more information technology assets, the given one of the second set of one or more artificial intelligence workloads comprising training a given artificial intelligence model, the at least one predicted power consumption associated with the execution of the given one of the second set of one or more artificial intelligence workloads on the one or more information technology assets comprising a per training epoch power consumption prediction for each training epoch of training the given artificial intelligence model;
determining a training process for the training the given artificial intelligence model based at least in part on the per training epoch power consumption prediction for each training epoch of training the given artificial intelligence model;
and
controlling execution of the given one of the second set of one or more artificial intelligence workloads on at least a given one of the one or more information technology assets based at least in part on the at least one predicted power consumption and the determined training process for the training of the given artificial intelligence model;
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.

4 more elements

Dependencies are inferred from the claim text.