Floating-point intensive applications—from Hollywood to Wall Street—have for years been challenged to provide both great performance and exactly the same results from run to run. Factors such as run-time selectable and optimized code paths, non-deterministic parallelism, array alignment and even the underlying operating system may influence the results computed. Intel® Math Kernel Library (Intel® MKL) 11.0 now includes features that can help users obtain conditional bitwise reproducible (CBWR) results when calling Intel MKL functions from their applications. This video will educate Intel software tool users and programmers on how to use Intel MKL and the Intel compiler to balance performance with the reproducible results their applications require.
Using the Intel Math Kernel Library 11.0 and Compiler to Obtain Run-to-Run Reproducible Results