Instructions and logic to perform floating-point and integer operations for machine learning

摘要: One embodiment provides for a graphics processing unit to accelerate machine-learning operations, the comprising multiprocessor having single instruction, multiple thread (SIMT) architecture, execute at least one instruction; and first compute included within multiprocessor, instruction cause perform two-dimensional matrix multiply accumulate operation, wherein operation includes 32-bit intermediate product of 16-bit operands sum based on product.