MLPerf Inference v3.1: Record-Breaking ai Benchmarks with New LLM and Recommendation Benchmarks
MLPerf Inference’s latest release, v3.1, sets new standards in the field of ai testing with improved benchmarks and record participation.
Unprecedented Participation and Performance Improvement
This iteration of the benchmark suite has seen remarkable engagement, boasting over 13,500 performance results and a significant up to 40% improvement in overall performance.
A Diverse Range of ai Innovators
The impressive achievement is underscored by the diverse pool of 26 submitters and over 2,000 power results. Tech giants like Google, Intel, NVIDIA join newcomers Connect Tech, Nutanix, Oracle, and TTA in pushing the boundaries of ai innovation.
Measuring ai System Performance Across Deployment Scenarios
MLPerf Inference is a crucial benchmark suite that evaluates the speed of ai systems in executing models across various deployment scenarios. These scenarios span from cutting-edge generative ai chatbots to essential safety features in vehicles, such as automatic lane-keeping and speech-to-text interfaces.
Two New Benchmarks: LLM and Recommendation
“The submissions for MLPerf Inference v3.1 demonstrate a wide range of accelerators being developed to serve ML workloads,” says Mitchelle Rasquinha, co-chair of the MLPerf Inference Working Group.
“The current benchmark suite covers a broad spectrum of ML domains, and the inclusion of GPT-J significantly expands our understanding of generative ai performance. Users can benefit greatly from these results when selecting optimal accelerators for their respective applications.”
MLPerf Inference: Focused on Data Center and Edge Systems
The primary focus of MLPerf Inference benchmarks is datacenter and edge systems. The v3.1 submissions exhibit a range of processors and accelerators catering to various use cases in computer vision, recommender systems, and language processing.
Balanced Competition: Open vs. Closed Submissions
MLPerf Inference benchmarks feature both open and closed submissions in the performance, power, and networking categories. Closed submissions employ a unified reference model for fair comparison across systems, while open submissions allow participants to showcase their unique models.
As ai continues to play a pivotal role in shaping our world, MLPerf’s benchmarks serve as indispensable tools for assessing and driving the future of ai technology.
Discover detailed MLPerf Inference v3.1 results
Explore upcoming enterprise technology events and webinars brought to you by TechForge
