MLPerf Inference Benchmark

Abstract

Machine-learning (ML) hardware and software system demand is burgeoning. Driven by ML applications, the number of different ML inference systems has exploded. Over 100 organizations are building ML inference chips, and the systems that incorporate existing models span at least three orders of magnitude in power consumption and four orders of magnitude in performance; they range from embedded devices to data-center solutions. Fueling the hardware are a dozen or more software frameworks and libraries. The myriad combinations of ML hardware and ML software make assessing ML-system performance in an architecture-neutral, representative, and reproducible manner challenging. There is a clear need for industry-wide standard ML benchmarking and evaluation criteria. MLPerf Inference answers that call. Driven by more than 30 organizations as well as more than 200 ML engineers and practitioners, MLPerf implements a set of rules and practices to ensure comparability across systems with wildly differing architectures. In this paper, we present the method and design principles of the initial MLPerf Inference release. The first call for submissions garnered more than 600 inference-performance measurements from 14 organizations, representing over 30 systems that show a range of capabilities.

Building a Camera-based Smart Sensing System for Digitalized On-demand Aircraft Cabin Readiness Verification

“…Regarding the selection of AI-processors and software tools for the optimal deployment of DNNs, we have established criteria to fulfill the system's requirements and analyzed some remarkable systems based on the MLPerf Inference benchmark suite (Reddi et al, 2019).…”

“…To support the AI-processor scouting process we rely on MLPerf Inference (Reddi et al, 2019), which is a relevant benchmark suite in the machine learning community for measuring how fast machine learning systems can process inputs and produce results using a trained DNN model. It was designed with the involvement of more than 30 organizations as well as more than 200 machine learning engineers and practitioners to overcome the challenge of assessing machine learning-system performance in an architecture-neutral, representative, and reproducible manner, despite the machine learning ecosystem's many possible combinations of hardware, machinelearning tasks, DNN models, data sets, frameworks, toolsets, libraries, architectures, and inference engines.…”

NNBench-X

“…Thus, our work urges taking the whole software-hardware co-design solution as a benchmark object instead of benchmarking pure hardware designs by providing a fixed set of applications. The recently released MLPerf inference benchmark [44] includes an Open Division under the same motivation as our study, although they are a preliminary release and the rules of Open Division are immature. Compared to the immature rules in this preliminary release, our methodology provides a concrete interface to take the model compression techniques as the input and generate the compressed models as the output.…”

ANNETTE: Accurate Neural Network Execution Time Estimation With Stacked Models

“…MLPerf [22] is an attempt by over 30 organizations to create an industry-wide standard benchmark to assess the vast number of machine learning software and hardware combinations, while DAWNBench [23] is led by academia. MLPerf limits the problem space by defining a set of scenarios, datasets, libraries, frameworks, and metrics.…”