Back to results
US 2026/0230392 A1Published application
Inferring hardware architectures to perform workflows at scale
AssigneeNVIDIA Corporation
InventorsSiddha Ganju, Ryan Albright, Elad Mentovich, Gal Ashkenazi, William Andrew Mecham, Benjamin Goska, Jordan Levy, William Ryan Weese, Scott Millward
Life of this patent
Estimating the term…
Abstract
Systems and methods disclosed herein can be used to determine an architecture of a data center based on one or more criteria. A system generates a digital representation of a data center, simulates how the digitally represented data center would perform under a certain workflow, and determines which of the simulated data centers satisfies the one or more criteria.
Claims20
A system, comprising:
- one or more processors to:
- generate a digital representation of a physical data center in a digital simulation environment;
- provide, as input to a machine learning model, information about a workflow to be performed in the physical data center, the machine learning model trained using benchmark data for a plurality of processing jobs performed using combinations of hardware in the physical data center;
- execute, in the digital simulation environment, one or more simulations based in part upon one or more hardware architectures output from the machine learning model as being potentially optimal for running the workflow in the physical data center;
- analyze performance data for the one or more simulations to determine whether the performance data satisfies one or more architecture selection criteria;
- and
- perform one or more additional simulations using other potential hardware architectures, inferred by the machine learning model based in part upon the performance data for the one or more simulations, until the performance data for at least one of the simulations satisfies the one or more architecture selection criteria.
4 more elements
A simulation environment, comprising:
- one or more processing units to execute a plurality of simulations of a data center running an anticipated workflow, the plurality of simulations having different hardware architectures inferred using a machine learning model trained using benchmark data for other jobs previously performed using various combinations of hardware, the machine learning model to infer additional potential hardware architectures based in part upon performance data for the plurality of simulations until at least one hardware architecture is identified that satisfies at least one architecture selection criterion.
A method comprising:
- creating a virtual model representing a data center within a simulated digital environment;
- inputting, for processing by machine learning model, details regarding an operational process to be executed within a physical infrastructure, the machine learning model trained using one or more reference data associated with various tasks carried out utilizing different configurations of the data center;
- generating simulations within the simulated digital environment of one or more system configurations identified by the machine learning model as being potentially effective for executing the operational process in the data center;
- evaluating performance metrics generated from the simulations to determine whether the performance metrics meet one or more data center criteria;
- and
- generating further simulations employing one or more alternative data center configurations suggested by machine learning model based on one or more insights derived from the performance metrics until the performance metrics align with the data center criteria for data center selection.
3 more elements
Dependencies are inferred from the claim text.