The Adept Benchmark Suite

The Adept benchmarks are used to exercise specific aspects of a system to allow measurement of both the performance and the power usage of operations on a wide range of hardware platforms. Their primary purpose is not to measure pure performance in terms of runtime or energy consumption, but to develop an understanding of the power usage of a system. We have released micro and kernel level benchmarks, and a RAPL wrapper library to perform power measurement on supported Intel processors.

To learn more about the Adept Benchmark Suite, you can view our Benchmark flyer or read the Adept White Paper.

Micro benchmarks

The micro benchmarks include basic operations on scalar data types, branch & jump statements, function calls, I/O operations, inter-process communication, memory access, network transfer, thread and process management, and sleep.

The different implementations that are available are:

Serial

OpenMP parallel 

SIMD 

Assembler 

 

Kernel benchmarks

The kernel benchmarks are basic linear algebra (BLAS) operations (including dot product, matrix-vector and matrix-matrix multiplication), 2D and 3D stencil computations, a conjugate gradient solver, and file parsing.

Serial 

OpenMP parallel 

MPI parallel

UPC parallel 

 

RAPL

Intel’s Running Average Power Limit (RAPL) can be used to measure power/energy consumption. We have implemented a wrapper library to read the values of the RAPL counters on supported Intel processors, and provide an example of how to use the library in the Adept benchmarks.

RAPL wrapper library

RAPL in Adept benchmark