MLCommons recently announced the release of MLPerf Client v2.0, the latest version of its industry-standard benchmark suite for AI performance on personal computers. The tool is used to evaluate how laptops, desktops, and workstations perform when running AI workloads locally.
New Workload Highlights
This update introduces several new features to keep pace with the rapid evolution of AI hardware and software.
Image Generation Category
A new image generation category has been added, featuring the Flux.2 klein 4B model as an experimental test, significantly expanding the ability to evaluate generative visual capabilities.
Agentic AI Category
A new Agentic AI category has been introduced, tested through software engineering (SWE) Agent and Data Analyst Agent scenarios. It reports end-to-end performance while breaking down LLM inference and tool execution times.
LLM Inference Test Upgrade
The mandatory workload has been upgraded from Phi 3.5 mini instruct to Phi 4 Mini Instruct, with Qwen 3 8B introduced as an experimental test. A new mid-summary task has been added to the base tasks, with input prompts of approximately 4K tokens.
MLPerf Client v2.0 was developed in collaboration with AMD, Intel, Microsoft, NVIDIA, and multiple PC OEM vendors. It is now open source on GitHub, and developers can download it for free and contribute code.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接