Adept Benchmark Suite released!
Posted: 23 Sep 2015 | 14:59
After a couple of years of measuring and trying to understand the power and energy consumption of (parallel) software and hardware, we have now released one of the key tools that we've been using as part of this research: the Adept Benchmark Suite!
While measuring performance (ie time to solution) is well understood, doing the same for power or energy is much less straightforward and often hardware dependent. The Adept Benchmark Suite relies on third party power measurement (such as instrumentation of the hardware) to be in place. However, to get users started with initial experiments, we provide a library to use RAPL (Running Average Power Limit) counters on Intel processors to measure the power of CPUs and memory, as well as some example code on how to use this library within the Adept Benchmarks.
The Adept Benchmark Suite consists of a wide range of benchmarks inspired by high-performance embedded all the way to high-performance scientific computing. The benchmarks are designed to characterise the efficiency (both in terms of performance and energy) of computer systems, from the hardware and system software stack to the compilers and programming models. Care was taken to expose and (where at all feasible) eliminate overheads so that measurements can be as accurate as possible. We want to get to the core cost of different computations and this often relies on techniques to avoid the compiler or CPU being smarter than you would like them to be...
Power measurement example: CPU power trace of the 27-point stencil benchmark running on an Intel Core i5 4670k.
More information on the Adept Benchmark Suite can be found in our latest public deliverable, which describes the different benchmarks in more detail. You can download the benchmarks from EPCC’s GitHub: https://github.com/EPCCed
We would very much welcome any feedback or comments – please let us know if you have suggestions for improvements or any feature requests! Contact us on firstname.lastname@example.org or on Twitter via @adept_project!
The Adept project addresses the challenge of energy-efficient use of parallel technologies. Adept builds on the expertise of software developers from high-performance computing (HPC) to exploit parallelism for performance, and on the expertise of Embedded systems engineers in managing energy usage.