Abstract: One embodiment provides an apparatus comprising a memory stack including multiple memory dies and a parallel processor including a plurality of multiprocessors. Each multiprocessor has a single instruction, multiple thread (SIMT) architecture, the parallel processor coupled to the memory stack via one or more memory interfaces. At least one multiprocessor comprises a multiply-accumulate circuit to perform multiply-accumulate operations on matrix data in a stage of a neural network implementation to produce a result matrix comprising a plurality of matrix data elements at a first precision, precision tracking logic to evaluate metrics associated with the matrix data elements and indicate if an optimization is to be performed for representing data at a second stage of the neural network implementation, and a numerical transform unit to dynamically perform a numerical transform operation on the matrix data elements based on the indication to produce transformed matrix data elements at a second precision.

Abstract: An apparatus to facilitate optimization of a convolutional neural network (CNN) is disclosed. The apparatus includes optimization logic to receive a CNN model having a list of instructions and including pruning logic to optimize the list of instructions by eliminating branches in the list of instructions that comprise a weight value of 0.

Abstract: A processing system includes a task queue to serve as a circular buffer. Each record in the queue may include a status field and a task field. A producer thread in the processing system may determine whether the queue is full, based on the status field in the record at the tail of the queue. The producer may add a task to the queue in response to determining that the status field in the record at the tail of the queue marks that record as empty. A consumer thread may determine whether the queue is empty, based on the status field in the record at the head of the queue. The consumer may execute a pending task identified by the record at the head of the queue, in response to determining that the status field in the head record marks that record as full. Other embodiments are described and claimed.

Abstract: A database includes a vertex array that contains coordinates of a three-dimensional mesh, a face array that contains pointers to target coordinates in the vertex array, the target coordinates defining a polygon in the three-dimensional mesh, a free vertex container that identifies elements in the vertex array that do not contain vertices, and a free face container that identifies elements in the face array that do not contain pointers.

Abstract: A database includes a vertex array that contains coordinates of a three-dimensional mesh, a face array that contains pointers to target coordinates in the vertex array, the target coordinates defining a polygon in the three-dimensional mesh, a free vertex container that identifies elements in the vertex array that do not contain vertices, and a free face container that identifies elements in the face array that do not contain pointers.