Abstract: Apparatuses, systems, and techniques to reduce a size of neural networks. In at least one embodiment, a size of a neural network is reduced by at least removing one or more neurons of the neural network and adjusting one or more layers of the neural network to compensate for the removed one or more neurons.

Abstract: Aspects of the present invention are directed to computer-implemented techniques for performing data compression and conversion between data formats of varying degrees of precision, and more particularly for improving the inferencing (application) of artificial neural networks using a reduced precision (e.g., INT8) data format. Embodiments of the present invention generate candidate conversions of data output, then employ a relative measure of quality to identify the candidate conversion with the greatest accuracy (i.e., least divergence from the original higher precision values). The representation can be then be used during inference to perform computations on the resulting output data.

Abstract: Aspects of the present invention are directed to computer-implemented techniques for performing data compression and conversion between data formats of varying degrees of precision, and more particularly for improving the inferencing (application) of artificial neural networks using a reduced precision (e.g., INT8) data format. Embodiments of the present invention generate candidate conversions of data output, then employ a relative measure of quality to identify the candidate conversion with the greatest accuracy (i.e., least divergence from the original higher precision values). The representation can be then be used during inference to perform computations on the resulting output data.

Abstract: Aspects of the present invention are directed to computer-implemented techniques for performing data compression and conversion between data formats of varying degrees of precision, and more particularly for improving the inferencing (application) of artificial neural networks using a reduced precision (e.g., INT8) data format. Embodiments of the present invention generate candidate conversions of data output, then employ a relative measure of quality to identify the candidate conversion with the greatest accuracy (i.e., least divergence from the original higher precision values). The representation can be then be used during inference to perform computations on the resulting output data.

Abstract: A physics software development kit (PSDK) provides scalable physics content as a “vertical” that defines one or more physics simulations for a graphics asset in a graphics scene. The vertical and the graphics asset may be provided in a verticals library associated with the PSDK or generated using the PSDK. The PSDK integrates the vertical into an existing graphics application to generate physically-realistic graphics content. The vertical may be scaled by a user according to the capabilities of a computer system that executes the PSDK or, alternatively, may be scaled by the PSDK based on received hardware capabilities information. The PSDK selectively offloads the physics simulations associated with the vertical to a physics processing unit to optimize usage of processor resources. In addition, the PSDK provides a technique to extract a graphics asset based on an existing 3D model of the object.

Abstract: A method of operating a Linear Complementarity Problem (LCP) solver is disclosed, where the LCP solver is characterized by multiple execution units operating in parallel to implement a competent computational method adapted to resolve physics-based LCPs in real-time.

Abstract: A method of operating a Linear Complementarity Problem (LCP) solver is disclosed, where the LCP solver is characterized by multiple execution units operating in parallel to implement a competent computational method adapted to resolve physics-based LCPs in real-time.

Abstract: A Linear Complementarity Problem (LCP) solver is characterized by multiple execution units operating in parallel to implement a competent computational method adapted to resolve physics-based LCPs in real-time.

Abstract: A projected iterative descent method is used to resolve LCPs related to rigid body dynamics, such that animation of the rigid body dynamics on a display system occur in real-time.

Abstract: An efficient quasi-custom instruction set for Physics Processing Unit (PPU) is enabled by balancing the dictates of a parallel arrangement of multiple, independent vector processors and programming considerations. A hierarchy of multiple, programmable memories and distributed control over data transfer is presented.

Abstract: A Linear Complementarity Problem (LCP) solver is characterized by multiple execution units operating in parallel to implement a competent computational method adapted to resolve physics-based LCPs in real-time.

Abstract: A method of operating a Linear Complementarity Problem (LCP) solver is disclosed, where the LCP solver is characterized by multiple execution units operating in parallel to implement a competent computational method adapted to resolve physics-based LCPs in real-time.

Abstract: A projected iterative descent method is used to resolve LCPs related to rigid body dynamics, such that animation of the rigid body dynamics on a display system occur in real-time.