Benchmarking Survey
The current landscape of high-performance computing consists of many approaches that depend on architectural capabilities, system/compiler configurations, parallel schemes, etc. Each approach has different perspectives and techniques. In this work we survey the current state of HPC, categorize the most common approaches, and evaluate several kernels under different high-performance regimes. Our study shows that there exist configuration and application-specific parameters that can be leveraged to increase performance further from current state-of-the-art implementations.
Kernels
- GEMM
- FIR Filter
- SVM
- PCA