Performance Improvement with Non-Uniform Loads on SMT Processors

Previously, the researcher conducted a study to improve processor performance focusing on uniform workloads. We suspect that the processor does not work optimally because there are several units that are inactive when working on a task. In this study, we want to show the performance of a computer if it is given a non-uniform load on the Intel Core i7-4790 and AMD Ryzen 5 1500X, so we can see the maximum performance of the processor because of the non-uniform workload. We use C and OpenMP to create a program code which is a matrix multiplication program m x n. Then, the researcher gives a different workload (integer and floating-point workload) into the program structure. We also provide additional workloads with rounds 5, 10, 15, 20 and 25 times. Finally, after each workload looped, we add para-loop to repeat both workloads 24 times simultaneously. The researcher executes the program 8 times per thread to get the best execution time. We found that in Intel Core i7 4790, the best speedup of the processor was 4.67 using 8 threads. Then, in AMD Ryzen 5 1500X, the best speedup of the processor was 5.21 using 8 threads. Just like speedup, the best IPC from the processor Intel Core i7 4790 is 7.00 when using 8 threads. In AMD Ryzen 5 1500X, the best IPC is 7.20 when using 8 threads. That means the speedup and IPC are directly proportional.