Researchers from MIT have invented a machine-learning tool that can precisely predict how long it takes given code to execute on a computer chip, which can help programmers tweak the code for better performance. With current computer machines, it is tough to develop a performance model to evaluate this throughput. The puzzle with building this analytical model for the complicated machines is that it is a tiresome process to build, it is error-prone and must be built from scratch for each processor production of the computing. Researchers believe the tool is a big step forward toward fully modeling the performance of these chips for enhanced productivity.