AmazonGranted patentAI & softwareInterest 4/10

Amazon patents system to reserve GPU cloud capacity in advance

Amazon describes a method for cloud users to reserve specific blocks of computing power, specifically those with graphics processing units (GPUs), for a set period. The system uses a pre-computed schedule to match user requests for a certain number of instances and time duration with available capacity blocks. Once reserved, the user can launch their instances within that time window, ensuring they are placed in locations that allow for fast communication between them.

Why it mattersThis is interesting for businesses that need consistent, high-performance computing power for tasks like AI training, as it guarantees they will have the specific hardware they need when they need it.

In the filing’s own words

Techniques for implementing and utilizing capacity cluster resource reservations are described. A managed compute service receives a user's request to identify a capacity block for use in launching compute instances. A schedule with pre-computed blocks, each corresponding to an amount of instances and an amount of time, is used to identify a block satisfying the request's criteria. The user can later obtain the capacity block and launch instances into the reservation during its time window, where placement rules associated with the reservation ensure the instances are hosted in locations enabling low-latency intercommunications.

What exactly they claim (claim 1)

1. A computer-implemented method comprising: generating, by a managed compute service of a cloud provider network, a schedule including a plurality of blocks of compute capacity hosted by the managed compute service that are available to be reserved by users of the managed compute service, wherein each of the blocks corresponds to compute capacity for a number of compute instances, of a type providing access to graphics processing unit (GPU) processing resources, for a window of time; receiving, at the cloud provider network, a request originated on behalf of a user to find a capacity block, the request identifying a desired number of compute instances and an availability duration for the desired number of compute instances; identifying, by the managed compute service based on use of the schedule, at least a first block of the plurality of blocks as providing compute capacity for the desired number of compute instances for the desired amount of time; transmitting, by the managed compute service, a response to the request that identifies at least the first block, the response including an offering identifier associated with the first block; receiving a request to obtain the first block as a first capacity block for the user, the request including the offering identifier; at a beginning of the window of time corresponding to the first capacity block, updating a data store to change an ownership of the compute capacity for the first block to be associated with an account of the user; and launching one or more compute instances on behalf of the user using the compute capacity.

Facts

Company
Amazon (Amazon Technologies, Inc.)
Official title
Capacity cluster resource reservations in a cloud provider network
Patent number
US12792420B2
Application number
18497806
Filed
Oct 30, 2023
Granted
Nov 10, 2026
Inventors
10
Claims
20

Classified under

G06F 9/5072 Physics & computingG06F 9/5038 Physics & computingG06F 9/5044 Physics & computing

More from Amazon in AI & software

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.