We have entered the era of application-driven architecture specialization, where significant power/energy/performance benefits are obtained by leveraging specialization of compute units, memory and interconnect - broadly referred to as accelerator fabrics. The majority of recent publications in architecture conferences addresses various aspects of the design, evaluation and deployment of these fabrics. However, the shift in research from general purpose computing to accelerator fabrics has not yet coalesced the architecture community around specific sets of benchmarks for key emerging application domains. There is some maturation of benchmarks in a few specific domains such as machine learning, but many others such as security/privacy, recommendation systems, graph-based and web-based applications, etc., are still lacking a commonly accepted set of benchmarks, thus slowing research progress.
The workshop will bring together researchers who have embarked upon new benchmarking efforts over the past several years. The goal of the workshop is to allow benchmarking researchers to share lessons learned and best practices for identifying, developing, and gaining adoption of benchmarks, and the broader community to learn about recent offerings that support architecture research. Ultimately, we seek to promulgate best practices in engaging with the prospective user communities for new benchmarks.
The program consists of presentations on individual benchmark suites selected among the submitted proposals followed by an opening panel discussing the topics above.