US 12,332,816 B1
Dynamic assignment of bus bandwidth for sending tensors to neural processing units
Je Ik Choi, Pyeongtaek-si (KR)
Assigned to DEEPX CO., LTD., Seongnam-si (KR)
Filed by DEEPX CO., LTD., Seongnam-si (KR)
Filed on Jan. 25, 2025, as Appl. No. 19/037,189.
Claims priority of application No. 10-2024-0114593 (KR), filed on Aug. 26, 2024.
Int. Cl. G06F 13/18 (2006.01); G06F 13/16 (2006.01)
CPC G06F 13/18 (2013.01) [G06F 13/1668 (2013.01)] 20 Claims
OG exemplary drawing
 
11. A method comprising:
determining whether a data starvation has occurred or is predicted to occur in at least one processing core configured to perform computation operations of at least one neural network model associated with tensors;
determining a priority of memory access operations of each of the tensors via a bus circuit responsive to determination of whether the data starvation has occurred or is predicted to occur in the at least one processing core;
assigning bandwidth of the bus circuit for sending each of the tensors to the at least one processing core according to the determined priority; and
sending each of the tensors to the at least one processing core via the bus circuit according to the bandwidth of the bus circuit assigned to send each of the tensors to the at least one processing core.