Performance and Power: Systematic Evaluation of AI Workloads on Accelerators with CARAML
This work addresses the need for standardized benchmarking of AI accelerators for researchers and practitioners, though it is incremental as it builds on existing benchmarking approaches.
The paper tackles the problem of evaluating AI accelerator performance and energy efficiency by introducing the CARAML benchmark suite, which systematically assesses transformer-based large language models and computer vision models on hardware from NVIDIA, AMD, and Graphcore, providing automated and reproducible measurements.
The rapid advancement of machine learning (ML) technologies has driven the development of specialized hardware accelerators designed to facilitate more efficient model training. This paper introduces the CARAML benchmark suite, which is employed to assess performance and energy consumption during the training of transformer-based large language models and computer vision models on a range of hardware accelerators, including systems from NVIDIA, AMD, and Graphcore. CARAML provides a compact, automated, extensible, and reproducible framework for assessing the performance and energy of ML workloads across various novel hardware architectures. The design and implementation of CARAML, along with a custom power measurement tool called jpwr, are discussed in detail.