Inventors:
Thomas H. Kong - Los Altos CA, US
Shaun Ho - Los Altos CA, US
Matthew Papakipos - Palo Alto CA, US
Assignee:
NVIDIA Corporation - Santa Clara CA
International Classification:
G06T 1/00
G06K 9/72
G06F 7/00
Abstract:
A computing system has a graphics processor, a graphics memory, main memory, a bridge, and a central processing unit configured to process floating-point data of a first fixed size. An interconnect grid includes communication paths to link the graphics processor, the graphics memory, main memory, the bridge, and the central processing unit. A computing system component (e. g. , the graphics processor or central processing unit) converts floating-point data to graphics floating-point data with a fixed size smaller than the fixed size of the floating-point data. The computing system passes the floating-point data and/or the graphics floating-point data over at least a portion of the interconnect grid. Alternately, the graphics processor may directly read and process previously compressed and stored graphics floating-point data.