Evaluation of Executing DGEMM Algorithms on Modern Multicore CPU

Pawel Gepner, Victor Gamayunov, and David L. Fraser


Algorithm design, Parallel computing, Performance evaluation, Benchmarking, Multi-core CPU


In our study we compared two types of algorithms for dense matrix-multiplications utilizing the latest Intel multicore CPU’s. We have been focused on the serialized version of the code implementing brute-force algorithm also known as the Cauchy algorithm versus Strassen implementation, the reference calculation was done with Intel Math Kernel Library (MKL). We investigated different types of the instruction set and looked at the implications to the performance. We also tested the accuracy for various algorithms, each of which having different scaling properties and effective use of the resources. Finally we proposed the effective hybrid method for Strassen algorithm implementation.

Important Links:

Go Back