Hardware performance measurement is evolving to keep pace with the massive integration of artificial intelligence into everyday devices. Primate Labs has launched Geekbench 7, a version that moves away from traditional synthetic tests toward concrete usage scenarios, reflecting a computational paradigm where AI and machine learning are ubiquitous.

GPU: Moving Beyond Gaming to Content Creation

The GPU benchmark has been fundamentally restructured. It is no longer just about counting polygons, but evaluating the ability to handle AI inference and multimedia production. New tests include real-time face tracking for social filters, image upscaling via super resolution, and background blurring for video conferencing.

The suite now covers advanced editing processes such as RAW image processing, LUT-based video color grading, and fluid simulations. A critical update is the addition of support for Nvidia CUDA, joining Metal, Vulkan, and OpenCL to allow more accurate analysis of Nvidia GPUs in professional environments.

CPU and Multi-Core Optimization

The approach to CPU calculation has been revised to prevent skewed results. Geekbench 7 now distributes workloads across multiple threads only when the real-world task typically does so; for instance, the HTML5 browser test has been removed from the multi-core suite since browsers are primarily single-threaded.

On the media front, specific tests have been introduced for AV1 video encoding (ideal for screen-sharing), Opus audio compression, and the use of the Whisper model for real-time captioning. To ensure greater precision, Primate Labs has expanded its datasets to include larger source code archives and complex technical PDF documents.

New Reference Standards

The new version also introduces a shift in hardware references, adopting the Ryzen 7 7700 as the baseline system. This update makes the tool essential for evaluating new architectures, including recent data center evolutions like Nvidia's Vera CPU, where AI agent orchestration requires highly sophisticated thread management.