US 12,079,895 B2
Systems and methods for disaggregated acceleration of artificial intelligence operations
Christian Markus Petersen, Golden, CO (US); and Narsing Krishna Vijayrao, Santa Clara, CA (US)
Assigned to Meta Platforms, Inc., Menlo Park, CA (US)
Filed by Meta Platforms, Inc., Menlo Park, CA (US)
Filed on Nov. 17, 2021, as Appl. No. 17/529,080.
Prior Publication US 2023/0153938 A1, May 18, 2023
Int. Cl. G06T 1/20 (2006.01); G06F 18/214 (2023.01); G06F 18/2431 (2023.01)
CPC G06T 1/20 (2013.01) [G06F 18/214 (2023.01); G06F 18/2431 (2023.01)] 20 Claims
OG exemplary drawing
 
1. A system comprising:
a disaggregated artificial intelligence (AI) operation accelerator comprising:
a dense AI operation accelerator configured to accelerate dense AI operations;
a sparse AI operation accelerator, physically separate from the dense AI operation accelerator, configured to accelerate sparse AI operations;
a high-bandwidth bus that directly communicatively couples the dense AI operation accelerator and the sparse AI operation accelerator; and
a scheduler comprising:
a receiving module that receives an AI operation, the AI operation comprising a task associated with an AI model;
an identifying module that identifies the AI operation as at least one of a dense AI operation or a sparse AI operation; and
a directing module that dynamically disaggregates AI operations into sparse operations and dense operations by directing:
the dense AI operation accelerator to accelerate the AI operation when the identifying module identifies it as a dense AI operation; and
the sparse AI operation accelerator to accelerate the AI operation when the identifying module identifies it as a sparse AI operation; and
a physical processor that executes the receiving module, the identifying module, and the directing module.